- Email: [email protected]
Hierarchical motion control for real time simulation of industrial robots
Available online at www.sciencedirect.com Available online at www.sciencedirect.com
ScienceDirect ScienceDirect Available online atonline www.sciencedirect.com Procedia CIRP 00 (2019) 000–000 Available at www.sciencedirect.com
ScienceDirect ScienceDirect
Procedia CIRP 00 (2019) 000–000
www.elsevier.com/locate/procedia www.elsevier.com/locate/procedia
Procedia CIRP 00 (2017) 000–000 Procedia CIRP 81 (2019) 713–718
52nd CIRP Conference on Manufacturing Systems 52nd CIRP Conference on Manufacturing Systems
www.elsevier.com/locate/procedia
Hierarchical motion control for real time simulation of industrial robots 28th CIRP Design for Conference, May simulation 2018, Nantes, France Hierarchical motion control realatime of industrial robots a Tadele Belay Tuli *, Martin Manns a a Tadele Belay *,9-11, Martin Manns Siegen,Paul-Bonatz-Str. 57076 Siegen, toUniversität analyze theTuli functional andGermany physical
A new methodology architecture of Siegen,Paul-Bonatz-Str. 9-11, 57076 Siegen, Germany * Corresponding author. Tel.: +49-271-740-2267; fax: +49-271-740-2542. E-mail address: [email protected] existing products for Universität an assembly oriented product family identification a a
* Corresponding author. Tel.: +49-271-740-2267; fax: +49-271-740-2542. E-mail address: [email protected]
Abstract Abstract
Paul Stief *, Jean-Yves Dantan, Alain Etienne, Ali Siadat
École Nationale Supérieure d’Arts et Métiers, Arts et Métiers ParisTech, LCFC EA 4495, 4 Rue Augustin Fresnel, Metz 57078, France
Multi axis machine tools implement real time motion control algorithms. Objects including difficult to manufacture and on the object surface. Force compliant industrial robots Multi axis machine tools implement real time motion control algorithms. Objects including difficult to manufacture and are used to manipulate deformable objects in real time simulation. In this research, a hierarchical method of motion controlling deformable shapes show variable properties when tools apply a pressure on the object surface. Force compliant industrial robots strategy is presented. The proposed approach is described regarding performance characteristics such as accuracy, repeatability, are used to manipulate deformable objects in real time simulation. In this research, a hierarchical method of motion controlling controllability of the motion segmentation and time dynamics. The result shows that a hierarchical control approach can be strategy is presented. The proposed approach is described regarding performance characteristics such as accuracy, repeatability, Abstract considered as a potential candidate to manipulate deformable objects where force requirements are critical. controllability of the motion segmentation and time dynamics. The result shows that a hierarchical control approach can be Inconsidered today’s business environment, the trend towards more product variety andwhere customization is unbroken. are Duecritical. to this development, the need of as a potential candidate to manipulate deformable objects force requirements *deformable Correspondingshapes author. Tel.: 3 87 37 54 30; E-mailwhen address: [email protected] show+33 variable properties tools apply a pressure
© 2019 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license agile andThe reconfigurable production systems Ltd. emerged to cope with various products and product families. To design and optimize production © 2019 Authors. Published by Elsevier (http://creativecommons.org/licenses/by-nc-nd/3.0/) systems as well as to choose the optimal product matches, product analysis areCC needed. Indeed, most of the known methods aim to © 2019 The Authors. Published by Elsevier Ltd. This is license an open access articlemethods under the BY-NC-ND license This is an open access article under the scientific CC BY-NC-ND (http://creativecommons.org/licenses/by-nc-nd/3.0/) Peer-review under responsibility of the committee of the 52nd CIRP Conference on Manufacturing Systems. analyze a product or one product family on the physical level. Different product families, however, may differ largely in terms of the number and (http://creativecommons.org/licenses/by-nc-nd/3.0/) Peer-review under responsibility of the scientific committee of the 52nd CIRP Conference on Manufacturing Systems. nature of components. This fact impedes an efficient comparison and choice of appropriate product familySystems. combinations for the production Peer-review under responsibility of the scientific committee of the 52nd CIRP Conference on Manufacturing Keywords: motion planning; industrial robot; realtime simulation; hierarchical control system. A new methodology is proposed to analyze existing products in view of their functional and physical architecture. The aim is to cluster Keywords: motion planning; industrial robot;product realtimefamilies simulation; control of existing assembly lines and the creation of future reconfigurable these products in new assembly oriented forhierarchical the optimization assembly systems. Based on Datum Flow Chain, the physical structure of the products is analyzed. Functional subassemblies are identified, and a1.functional analysis is performed. Moreover, a hybrid functional and physical (HyFPAG) is the whichsystem depictsand the Background adapt architecture tasks usinggraph learning skills in an output automated similarity between product families by providing design support to both, production system planners and product designers. An illustrative apply multi-level control strategies [6–8]. 1. Background tasks using skills in an automated example of a nail-clipper is used to explain the proposed methodology. Anadapt industrial case studylearning on two product families of steeringsystem columnsand of In multi axes machine tools, objects including difficult to Modern machine tools are capable apply of multi-level control strategies [6–8].to generate tool path thyssenkrupp Presta France is then carried out to give a first industrial evaluation the proposed approach. and machine deformable shapes manipulated using points for machine free forms (e.g. [9], tool complex ©manufacture 2017 The Authors. Publishedtools, by Elsevier B.V. In multi axes objectsare including difficult to Modern tools are turbine capable blades) to generate path real time motion control. Industrial robots as intelligent geometries (e.g. additively manufactured soft-grippers) [10], Peer-review under the scientific committee of the 28th CIRP Design 2018.(e.g. turbine blades) [9], complex manufacture andresponsibility deformableof shapes are manipulated using points forConference free forms
manufacturing toolscontrol. are adapted to new geometric variants real time motion Industrial robots as intelligent in automated manufacturing tools are adapted to new geometric variants production system. The wide range of industrial and mechanical properties using sensor models inapplications automated makes robot motion planning and controlling strategies costly production system. The wide range of industrial applications and complex [1]. Recently, robot motion control is being makes robot motion planning and controlling strategies costly 1.simplified, Introduction and robots are sharing a common working space and complex [1]. Recently, robot motion control is being with humanandbeings [2,3].sharing Some aautomated manufacturing simplified, common working space Due to that therobots fast are development in the domain of processes require low payloads use force compliant with human beings [2,3]. Some automated manufacturing communication and an ongoing trend of digitization and robots. Particularly, increase of force use compliant industrial processes thatmanufacturing require low enterprises payloads force compliant digitalization, are facing important robots with an open architecture, has created a common robots. Particularly, increase ofenvironments: force compliant industrial challenges inboth today’s market a continuing platform for research and manufacturing communities to robots with an open architecture, has created a times common tendency towards reduction of[4]. product development and exchange practical use cases Open architectures for robot platform for both research and manufacturing communities to shortened product Infreedom addition, there is an increasing controllers providelifecycles. scientific toarchitectures implement advanced exchange practical use cases [4]. Open for robot demand of customization, being atprocesses the same[5]. time in a global control laws into manufacturing Direct access controllers provide scientific freedom to implement advanced competition with competitors all over the world. This trend, to low level joint systems such as servo systems and joint controlislaws into manufacturing processes [5]. Direct access which inducing the development from macro to micro kinematics invites to optimize motion planning techniques, to low level joint suchlotassizes servodue systems and joint markets, results in systems diminished to augmenting kinematics invites to optimize motion planning techniques, product varieties (high-volume to low-volume production) [1].
and surface (e.g. forming (e.g. robotic incremental forming) [11]. geometries additively manufactured soft-grippers) [10], However, soft and deformable objects such as additively and surface forming (e.g. robotic incremental forming) [11]. manufactured are difficult to manipulate due to the However, softelastomers and deformable objects such as additively unknown behavior of the material deformation when pressure manufactured elastomers are difficult to manipulate due to the is applied [12,13]. Force sensor model approaches are unknown behavior of the material deformation when pressure of the product range and characteristics manufactured and/or presented by various groups to control motions of industrial is appliedin [12,13]. Force sensor model approaches assembled this system. In this context, the main challengeare in robots to manipulate such objects [14,15].motions presented by various groups to control of industrial modelling and analysis is now not only to cope with single To implement motion algorithms, motion features robots to manipulate such control objects [14,15]. products, a limited product range or existing product families, such as execution time, position accuracy, control strategy, To implement motion control algorithms, motion to features but also to be able to analyze and to compare products define collision avoidance and error interpretations are mainly such as execution time, position accuracy, control strategy, new product families. It can be observed that classical existing considered. Motion control is used to prescribe are a robot to collisionfamilies avoidance and error interpretations mainly product are regrouped in within function of defined clients ortime features. follow a desired trajectory a by considered. Motionoriented controlproduct is usedfamilies to prescribe a robot to However, assembly are hardly to find. complying to force ortrajectory torque requirements. There are time multiple follow a desired within a defined by On the product family level, products differ mainly in two motion control approaches including intelligent based complying to force or(i)torque requirements. Therevision are multiple main characteristics: the number of components and (ii) the control strategies [16], multi heterogeneous sensor motion control approaches including intelligentelectronical). vision based based type of components (e.g. mechanical, electrical, control strategies [17], reality based control strategies control strategies [16],virtual multi heterogeneous sensor based Classical methodologies considering mainly single products control strategies [17], virtual reality based control strategies or solitary, already existing product families analyze the 2212-8271 © 2019 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license To cope with this augmenting variety as well as to be able to product structure on a physical level (components level) which (http://creativecommons.org/licenses/by-nc-nd/3.0/) 2212-8271 © 2019 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license an efficient definition and identify possible optimization potentials in the existing causes regarding Peer-review under responsibility of the scientific committee of the 52nd CIRP Conference on difficulties Manufacturing Systems. (http://creativecommons.org/licenses/by-nc-nd/3.0/) production system, it is important to have a precise knowledge comparison of different product families. Addressing this and mechanical using sensor models Keywords: Assembly;properties Design method; Family identification
Peer-review under responsibility of the scientific committee of the 52nd CIRP Conference on Manufacturing Systems.
2212-8271 © 2019 The Authors. Published by Elsevier Ltd. This is an©open article Published under theby CC BY-NC-ND 2212-8271 2017access The Authors. Elsevier B.V. license (http://creativecommons.org/licenses/by-nc-nd/3.0/) Peer-review under responsibility of scientific the scientific committee theCIRP 52ndDesign CIRPConference Conference2018. on Manufacturing Systems. Peer-review under responsibility of the committee of the of 28th 10.1016/j.procir.2019.03.181
714 2
Tadele Belay Tuli et al. / Procedia CIRP 81 (2019) 713–718 Tadele Belay Tuli and Martin Manns / Procedia CIRP 00 (2019) 000–000
[18], geometrically constrained motion control, and fuzzy based force/position hybrid control [15]. Today, as digital manufacturing concepts have been widely adopted, industrial robots are available with an open controller architecture. Converting ideas into reality is being further simplified. Real time motion planning and control is effectively presented for handling or grasping of rigid objects where object deformation can be negligible. Handling of deformable objects where the amount of applied pressure is critical e.g. hollow micro structures are still open for further investigations. In such cases, a real time adaptable motion generation approach is required. Deformable objects are difficult to handle; first they are highly sensitive to applied force, second a particle of the object changes a position with a small applied pressure. Third, controlling a force at a desired value is difficult [10,13,19–21]. Robot programming system (ROS) is a meta-operating system which provides the desired service of a dedicated robot for hardware in the loop simulation. The operating system includes hardware abstraction, low level control of sensors and actuators, implementation of functionality share, messaging, and package management. In manufacturing automation, robots have some unique functional requirements, job specification, working environment and operation parameters unlike service robots. The initiative of Yaskawa Motoman Robotics, Southwest Research Institute, and Willow Garage, has extended the functionality of ROS into the industrial community with the principal goal of manufacturing automation and robotics. ROS-I runs on the Linux operating system, and it does not provide assurance to robot motion. To overcome this issue, robot designers sub-divided the system into real time and non-real time subsystems. The real time robot motion execution is implemented using embedded system coupling of ROS-Industrial and ROS bridge [22]. The Universal Robot UR5 that are considered in this research is a low payload type industrial robots with six revolute joint variables. It is known to be among the first generation of robots to be used in human robot collaboration. It supports a TCP/IP communication at 100Mbyte [23]. ROS-I provides subscriber and publisher functionality to receive and send controller parameters [4]. 2. Objective In this work, we propose a hierarchical controller approach to control a motion of an industrial robot at real time. The approach considers tool center point force property and joint space variables to derive a robust motion control that hierarchically adapts to the level of object deformation. Additively manufactured elastomer cubes with hollow cross sections are chosen to develop a real time force control. We investigate how to apply pressure on such a deformable object that has a resilient motion property. In this case, a micro motion movement is considered to characterize an applied force. In this regard, we justify that in a real time simulation, acceptable ranges of target force can be achieved by implementing PID controllers into the system. Finally, we verify our approach by implementing a real time simulation using ROS-I. The Python programming language is used to
establish system interface. We develop a function that is used for data acquisition, data analysis and data transmission using ROS-I’s built in APIs . 3. Robot motion modeling Given the initial and final joint space variables such as position, forward robot kinematics computes the global position and orientation of the robot tool center point with respect to a fixed reference frame. Alternatively, inverse kinematics maps the Cartesian space of the robot end effector to the joint space variables. The analytical kinematic model presented in [24] is adopted. Considering the kinematic configuration of six axes of the considered robot, the joint state position vector of wrist three (i.e. frame six) with respect to the base frame (i.e. frame zero) can be given in equation 1. 𝑞𝑞1 (𝑡𝑡) 𝑞𝑞2 (𝑡𝑡) 𝑞𝑞(𝑡𝑡) = 𝑞𝑞3 (𝑡𝑡) ⋮ [ 𝑞𝑞6 (𝑡𝑡) ]
(1)
Where 𝑞𝑞(𝑡𝑡) is the joint state position vector and 𝑞𝑞1 (𝑡𝑡), 𝑞𝑞2 (𝑡𝑡), 𝑞𝑞3 (𝑡𝑡), . . . , 𝑞𝑞6 (𝑡𝑡) are joint state variables of six axis. 3.1. Motion characterization In real time motion control of a robot, desired joint positions are expected at a specific time frame. The joint space variables are broken down into N small segments along the motion axis at a time interval of Ti. The time dynamics of joint variable is set to the minimum actuation time. Considering v, the maximum velocity of the robot and Tmin, the minimum actuation time. The time step used to characterize the motion segment can be computed as; 𝑇𝑇𝑚𝑚𝑚𝑚𝑚𝑚 = min(𝑇𝑇𝑞𝑞𝑞𝑞 )
(2)
where 𝑇𝑇𝑞𝑞𝑞𝑞 is a joint actuation time of joint i. From equation 2, we can express an infinitesimal path segment Δ𝑢𝑢 and actuation time step mathematically as Δ𝑢𝑢 = 𝑣𝑣𝑇𝑇𝑚𝑚𝑚𝑚𝑚𝑚 . In this case, the real time motion generation expects either Tmin or Δ𝑢𝑢 to be constant to maintain a motion that is commanded by velocity. The step size Δ𝑢𝑢 is limited to the discrete servo control interval (i.e. 𝑇𝑇𝑖𝑖 ≥ 𝑇𝑇) , where 𝑇𝑇 is the closure time. Given the joint position 𝑞𝑞𝑖𝑖 (𝑡𝑡), step size (Δ𝑢𝑢𝑖𝑖 ) and step size N can be given as; Δ𝑢𝑢𝑖𝑖 (𝑡𝑡) =
𝑞𝑞𝑖𝑖 (𝑡𝑡) N
(3)
In section 4.3, we characterize joint space variables in order to get smooth, robust and controllable motion profiles by applying equation 3. This approach is important to simplify real time motion generation.
Tadele Belay Tuli et al. / Procedia CIRP 81 (2019) 713–718 Tadele Belay Tuli and Martin Manns / Procedia CIRP 00 (2019) 000–000
715 3
3.2. Motion planning
3.3. Motion Rescaling/adaptation
A trapezoidal velocity profile is a well-known approach for motion planning in the communities of robotics and automated machine tools. However, the ideal trapezoidal velocity profile produces undesired motion behavior at the discontinuities of the velocities. This will impose perturbation to the force controller. To avoid this, we increase the step size (N) and implement a parabolic blending at the discontinuities. Considering N path segments, an intermediate discrete segment k can be modified using time derivative of joint space variables. [16] presented a mathematical procedures to be followed to remove jumps and discontinuities using differential equations. At time instant tk, the time differential of joint space qk(t) possesses a parabolic profile around tk.
In equation 6, there is a piecewise function that interpolates time to ramp up, move constant and ramp down. In a real time implementation of the algorithm, maintaining a constant time for each motion segment is difficult. Instead, we apply a constant segmentation throughout the initial and final position. The discrete approach that maps the tool center position and joint variable at infinitesimal interval is given in equation 7 – 9 by coupling to the robot kinematics (see section 4).
𝑞𝑞̇ 𝑘𝑘−1,𝑘𝑘 = 𝑞𝑞̈ 𝑘𝑘−1,𝑘𝑘 =
𝑞𝑞𝑘𝑘 −𝑞𝑞𝑘𝑘−1
(4)
𝑞𝑞̇ 𝑘𝑘 −𝑞𝑞̇ 𝑘𝑘−1
(5)
Δ𝑡𝑡𝑘𝑘−1 Δ𝑡𝑡𝑘𝑘−1
Where, Δ𝑡𝑡𝑘𝑘 = 𝑡𝑡𝑘𝑘 − 𝑡𝑡𝑘𝑘−1 is the corresponding time of the distance 𝑞𝑞𝑘𝑘 and 𝑞𝑞𝑘𝑘−1 . Accordingly, Δ𝑡𝑡𝑘𝑘−1 ,𝑘𝑘 denotes the time interval to generate a trajectory of 𝑞𝑞𝑘𝑘 and 𝑞𝑞𝑘𝑘−1 .
𝑁𝑁
Case i: Ramping up (i.e. 𝑘𝑘1 ≤ ) 1
3
1
Δ𝑢𝑢1 = (𝑞𝑞𝑖𝑖 + 𝑎𝑎(𝑘𝑘1 Δ𝑡𝑡)2 ) N
2
Case ii: Constant velocity ( i.e.𝑘𝑘2 > 1
1
𝑁𝑁 3
Δ𝑢𝑢2 = ( (𝑞𝑞𝑓𝑓 + 𝑞𝑞𝑖𝑖 − 𝑣𝑣(𝑁𝑁Δ𝑡𝑡)) + 𝑣𝑣(𝑘𝑘2 Δ𝑡𝑡)) N
2
2
& 𝑘𝑘3 ≤ 𝑁𝑁) 3
2
Case iii: Ramping down ( 𝑘𝑘3 > 𝑁𝑁 𝑠𝑠 & 𝑘𝑘3 ≤ 𝑁𝑁) Δ𝑢𝑢3 =
1
N
1
3
ሺሻ
ሺͺሻ
1
(𝑞𝑞𝑓𝑓 − 𝑎𝑎(𝑁𝑁Δ𝑡𝑡)2 + 𝑎𝑎(𝑁𝑁Δ𝑡𝑡)(𝑘𝑘3 Δ𝑡𝑡) − (𝑘𝑘3 Δ𝑡𝑡)2 )ሺͻሻ 2
2
The effectiveness of equation 7 – 9 depends on the selection of sampling size (N) and time step (Δ𝑡𝑡). 4. Controller design Fig. 1. A trapezoidal velocity profile with a parabolic blend.
Implementing a linear segment parabolic blend inside a hierarchical control specifically, in which position and force are dependent on sensor models, it is important to observe the tool center mounted force behavior and characterize it. The algorithm is capable to reduce unwanted motion behaviors such as jerk and discontinuities by creating smooth motion transition. In this case, velocity is characterized in three levels. Initially there is “ramping up” for one third of the closure step time followed by a constant motion for the same time before it is “ramping down” to the desired position. The corresponding notation is summarized in equation 6 (see [19]).
𝑞𝑞(𝑡𝑡) =
1
{
2
1
𝑞𝑞𝑜𝑜 + 𝑎𝑎𝑡𝑡 2 , 𝑡𝑡 ∈ [0, 𝑡𝑡𝑏𝑏 ] 2
(𝑞𝑞𝑓𝑓 + 𝑞𝑞𝑜𝑜 − 𝑣𝑣𝑡𝑡𝑓𝑓 ) + 𝑣𝑣𝑣𝑣 , 𝑡𝑡 ∈ (𝑡𝑡𝑏𝑏 , 𝑡𝑡𝑠𝑠 ] 1
1
𝑞𝑞𝑓𝑓 − 𝑎𝑎𝑡𝑡𝑓𝑓2 + 𝑎𝑎𝑡𝑡𝑓𝑓 𝑡𝑡 − 𝑡𝑡 2 , 𝑡𝑡 ∈ (𝑡𝑡𝑠𝑠 , 𝑡𝑡𝑓𝑓 ] 2
2
ሺሻ
where 𝑣𝑣 , 𝑎𝑎 , 𝑡𝑡𝑎𝑎 and 𝑡𝑡𝑓𝑓 are the mean velocity, acceleration, initial blending time and final blending time respectively. In the same time, we can represent 𝑡𝑡𝑠𝑠 = 𝑡𝑡𝑓𝑓 − 𝑡𝑡𝑏𝑏 and initial and final joint variables using [𝑞𝑞𝑜𝑜 , 𝑞𝑞𝑓𝑓 ].
Parts that are additively manufactured using soft materials such as elastomers and silicon exhibit differently [10,21]. Particularly, it is difficult to use industrial robots to precisely manipulate such objects. It requires robust system that regulates joint space variables state by analyzing force behavior at the tool center point. In this section, requirements are identified to design a controller as it is stated below. 1. 2. 3.
Target force: The robot tool center point applies a pressure along a desired. Tool center point constraint: The robot tool center point has a discrete control points to limit unnecessary motions. Motion step: The motion has to be smooth. Therefore, the motion is characterized into features such as time, velocity and distance.
In addition, control parameters are defined at the high level system and low level subsystem. Target force is a low level controller input whereas a tool center position is an input at the high level controller. The desired outputs are controlled force and position. In section 4.1 and 4.2, control strategies for force and position control are presented.
Tadele Belay Tuli et al. / Procedia CIRP 81 (2019) 713–718 Tadele Belay Tuli and Martin Manns / Procedia CIRP 00 (2019) 000–000
716 4
4.1. Position based discrete control
•
Position controller is a discrete controller that is implemented to constrain the tool center point motion along a desired axis. The initial and final discrete position of the tool center point is specified to avoid robot collision. The main role of position control is to continuously generate a motion of the robot using a computed time to achieve the target motion. If the target force cannot be achieved in a desired working space, the controller will bypass the role of low level controller to avoid undesired motion or collision (refer Fig. 2).
•
Fig. 2. A hierarchical approach motion control.
PID (Proportional – Integral – Differential) control is a classical closed loop output feedback control that is represented mathematically by equation 10 (see [25]). It is used to prescribe how the control law acts. 𝑡𝑡
𝑘𝑘𝑑𝑑 𝑑𝑑𝑑𝑑(𝑡𝑡) 𝑑𝑑𝑑𝑑
𝑑𝑑𝑑𝑑
The list of materials and devices used in this experiment are: laptop (i.e. DELL precision 5520, 32GB RAM, Core i7 laptop), ATI Net F/T sensor, transducer cable (CAN), Ethernet hub, Ethernet to thunderbolt adaptor, customized end effecter and universal robot. Besides, system supporting tools such as Linux OS (Ubuntu 18.04 - Bionic), ROS (Melodic release), and Python 2.7.2 are configured to the hardware system.
Fig. 3. System configuration.
4.2. Force control
𝑢𝑢(𝑡𝑡) = 𝑘𝑘𝑝𝑝 𝑒𝑒(𝑡𝑡) + 𝑘𝑘𝑖𝑖 ∫0 𝑒𝑒(𝑡𝑡)𝑑𝑑𝑑𝑑 +
•
Workspace noises and disturbances are either not measured or not decoupled from the system. The errors exhibited due to incompatibility of the robot controller and the remote controller system are neglected. System isolation of independent processes is not considered.
ሺͳͲሻ
where kp, ki and kd represent the associated gains to the proportional, integral and derivative components respectively.
As Fig. 4 illustrates, the ROS-I system handles TCP/IP connections through the intermediate hub to communicate to robot joint sensors and external force sensor. In ROS-I there are topics that are used to channel joint state, transform information, trajectory action and wrench data through messages [4]. The force sensor mounted at tool center point uses only a subscriber function to collect force data whereas the communication between robot and ROS server is two directional (i.e. subscriber and publisher).
4.3. Software configuration ROS-I (Robot Operating Software Industrial) is a popular open source platform that can be used to implement feedback control strategies including multi input and multi output systems. However, industrial robots that are supported by ROS-I have to possess an open controller architecture. In ROS-I system, two packages are created to interface robot system and force sensor system. Both packages use a launcher that prompts for IP address to establish a connection. The specific topics will be instantiated through the corresponding node to handle messages. During the interface configuration, existing open source codes for universal robot [22] and force sensor [26] are used as a starting point. 5. Experimental set-up and real time implementation 5.1. Experimental set-up The experimental set up is implemented in our lab (see fig. 3). During the experiment:
Fig. 4. Hardware configuration and simulation for the tests.
5.2. Real time simulation For a real time simulation, a set of rules are abstracted. In these rules, static IP address is assigned to a robot and a force transducer in the same network (see Fig. 3). In ROS-I both systems are launched separately to enable the communication of the ROS-I system and hardware systems. A subscriber function is developed to extract position and force data into the main function to analyze the data. A publisher function sends a decision of the controller to the robot.
Tadele Belay Tuli et al. / Procedia CIRP 81 (2019) 713–718 Tadele Belay Tuli and Martin Manns / Procedia CIRP 00 (2019) 000–000
Deformable object that exhibits variable mechanical property when a force is applied is considered to analyze applicability of the proposed approach. Such use case is common in most industrial applications e.g. a force control for foam, cloth, rubber and other material. As it is shown in Fig. 5, target force is defined to apply a pressure on a deformable cube (i.e. additively printed silicon with a dimension of 20x20x20mm3). The applicability of hierarchical controller is analysed at a minimum motion segment and due to resilience of the object during force regulation, adaptive motion planning is proposed.
a.
717 5
holder and shop floor disturbance. The experiment is deliberately conducted in a shop floor (see Fig. 4) in which multiple machines are running (i.e. KUKA Jet, SCARA Bosch, and KUKA LWR). This will lead us to generalize that, the experimental setup can be replicated in an industrial environment. For real time simulation purpose, a force control is configured with an optimal controller gain parameters. The gains are sensitive to system noises. We observed that, for uncontrolled system variables, it is necessary to tune the PID gains to obtain better performance. In our system, we employed kp = 100, kd = 0.1 and ki =0 after optimization of the system.
b.
Fig. 5. Use case definition and workspace setup (F denotes the tool center force measured by external force sensor, and Δx(t) represents the time variable of object deformation).
5.3. Description of result Simulation results are described using performance characteristics such as accuracy, repeatability, motion smoothness, controllability of the motion segmentation and time dynamics. These criteria are used to relate the performance of robot movement, task accomplishment and difficulty level of the job with regard to the experimental setup that is discussed in section 5.1 and 5.2.
a.
b. Fig. 6. A force and time output for a target force of 30N using a hierarchical control law (a) Overall force behaviour (b) A cubic curve fitting for target preservation phase.
During a hardware in the loop simulation, noise is observed from the system configuration. The effect is shown by spikes particularly for a higher applied pressure. In our case, at 30N the robot motion is not smooth as the seeking phase (see Fig. 6). The source of this noise can be associated to robot joint stiffness, transfer system drive, work piece
Fig. 7. Residual error distribution of a PID controller output for gain parameters of (kp = 100, kd = 0.1 and ki =0)
6. Discussion In Fig. 6 – 7 the force characteristic is shown. There are three phases in the experiment. These are approach, seeking and target preservation. In each phase, there exist different time dynamics and motion speeds. In the first phase, the robot moves from its home position to the starting position of force measurement. In this case, the robot moves with 75% of the robot maximum speed. The second phase is characterized as force seeking motion. In this stage the robot force applies a pressure to seek a user defined target force. In this situation, the robot speed is analytically determined based on the motion planning algorithm. If the robot achieves the target force, the force controller will regulate the joint sensor to maintain the desired force. In our use case, the object is deformable and it does not maintain the measured force for some duration of time. Hence, it is necessary to apply a change in position to update the new force value to the target one. The simulation result shows, the force is controlled with an error of ± 0.5N. As it is explicitly shown in Fig. 6(b) spikes are observed in a random distribution. In general, the controller is capable to regulate the applied pressure a maximum of ±1N. An observed peak beyond ±0.5N is 6.67%. If we apply a constant pressure at a fixed robot joint position, the force value decreases gradually. This requires a controller action to update the force to a target force. In this research, potential challenges such as deformable object dynamics control and force control in a case of super slow motion is identified for future work. However, hierarchical control method can be considered as an initial approach to control the sensibility of applied pressure a case of deformable object manipulation for industrial applications.
718 6
Tadele Belay Tuli et al. / Procedia CIRP 81 (2019) 713–718 Tadele Belay Tuli and Martin Manns / Procedia CIRP 00 (2019) 000–000
7. Conclusion Real time motion control algorithms are being used for multi axes machine tools. Objects including difficult to manufacture and deformable shapes have dynamic properties when tools and end effecters apply pressure on the surface. This makes handling of deformable objects often difficult particularly in real time simulation. In this paper, an initial approach that uses a hierarchical motion control strategy is presented. Position control is implemented at a high level to constrain the tool center point motion within a defined working space. In addition, the position controller is proposed to generate joint state trajectories based on trapezoidal velocity profile, which blends the trajectory using parabolic equation. The force controller is a low level controller that regulates the applied force comparing with the target force. The real time simulation is implemented using the ROS-I platform and the Python programming language. Generic use case that satisfies deformability, resilient property and unknown mechanical property is identified and developed using additive manufacturing technology. The result indicates that, even though system disturbances have shown an influence in the smoothness of the data acquired, hierarchical motion controlling strategy has a potential to manipulate deformable objects where an applied tool pressure is critical. The study and comparison of the mechanical properties regarding shape, cross sectional area and mechanical characteristics including the system dynamics are to be considered as future work. References [1] Meyes R, Scheiderer C, Meisen T. Continuous Motion Planning for Industrial Robots based on Direct Sensory Input. Procedia CIRP 2018;72:291–6. doi:10.1016/j.procir.2018.03.067. [2] Mohammed A, Wang L. Brainwaves driven human-robot collaborative assembly. CIRP Annals 2018;67:13–6. doi:10.1016/j.cirp.2018.04.048. [3] Glogowski P, Lemmerz K, Hypki A, Kuhlenkötter B. ROS-Based Robot Simulation in Human-Robot Collaboration. In: Karafillidis A, Weidner R, editors. Developing Support Technologies: Integrating Multiple Perspectives to Create Assistance that People Really Want, Cham: Springer International Publishing; 2018, p. 237–46. doi:10.1007/978-3030-01836-8_23. [4] Koubaa A, editor. Robot Operating System (ROS): The Complete Reference (Volume 2). Springer International Publishing; 2017. [5] Tsardoulias E, Mitkas P. Robotic frameworks, architectures and middleware comparison. ArXiv:171106842 [Cs] 2017. [6] Vonásek V, Penc O, Přeučil L. Guided Motion Planning for Modular Robots. In: Hodicky J, editor. Modelling and Simulation for Autonomous Systems, Springer International Publishing; 2014, p. 217–30. [7] Meyes R, Tercan H, Roggendorf S, Thiele T, Büscher C, Obdenbusch M, et al. Motion Planning for Industrial Robots using Reinforcement Learning. Procedia CIRP 2017;63:107–12.
doi:10.1016/j.procir.2017.03.095. [8] Larsen L, Schuster A, Kim J, Kupke M. Path Planning of Cooperating Industrial Robots Using Evolutionary Algorithms. Procedia Manufacturing 2018;17:286–93. doi:10.1016/j.promfg.2018.10.048. [9] Altintas Y. Manufacturing Automation by Yusuf Altintas. Cambridge Core 2012. doi:10.1017/CBO9780511843723. [10] Manns M, Morales J, Frohn P. Additive manufacturing of silicon based PneuNets as soft robotic actuators. Procedia CIRP 2018;72:328–33. doi:10.1016/j.procir.2018.03.186. [11] Lu H, Liu H, Wang C. Review on strategies for geometric accuracy improvement in incremental sheet forming. Int J Adv Manuf Technol 2019. doi:10.1007/s00170-019-03348-3. [12] Shepherd R. Design of elastomeric composites for the additive manufacturing of robots. Composites at Lake Louise 2017 2017. [13] Otaduy MA, Tamstorf R, Steinemann D, Gross M. Implicit Contact Handling for Deformable Objects. Computer Graphics Forum 2009;28:559–68. doi:10.1111/j.1467-8659.2009.01396.x. [14] Kelmar L, Lumia R. Sensory interactive robot trajectory control using a real-time world model. Journal of Intelligent & Robotic Systems 1992;5:39–47. doi:10.1007/BF00357129. [15] Chaudhary H, Panwar V, Sukavanum N, Prasad R. Fuzzy PD+I based Hybrid force/ position control of an Industrial Robot Manipulator. IFAC Proceedings Volumes 2014;47:429–36. doi:10.3182/20140313-3IN-3024.00062. [16] Corke P. Vision-Based Control. In: Corke P, editor. Robotics, Vision and Control: Fundamental Algorithms In MATLAB® Second, Completely Revised, Extended And Updated Edition, Cham: Springer International Publishing; 2017, p. 537–63. doi:10.1007/978-3-319-544137_15. [17] Mozumdar P, Helal TS, Ababnah A, Akter L. Heterogeneous sensor deployment strategy based on optimal control theory. 2015 International Conference on Electrical Engineering and Information Communication Technology (ICEEICT), 2015, p. 1–6. doi:10.1109/ICEEICT.2015.7307446. [18] Burdea GC. Invited review: the synergy between virtual reality and robotics. IEEE Transactions on Robotics and Automation 1999;15:400–10. doi:10.1109/70.768174. [19] Kaboli M, Yao K, Cheng G. Tactile-based manipulation of deformable objects with dynamic center of mass. 2016 IEEE-RAS 16th International Conference on Humanoid Robots (Humanoids), 2016, p. 752–7. doi:10.1109/HUMANOIDS.2016.7803358. [20] Sanchez J, Corrales J-A, Bouzgarrou B-C, Mezouar Y. Robotic manipulation and sensing of deformable objects in domestic and industrial applications: a survey. The International Journal of Robotics Research 2018;37:688–716. doi:10.1177/0278364918779698. [21] Panetta J, Zhou Q, Malomo L, Pietroni N, Cignoni P, Zorin D. Elastic textures for additive fabrication. ACM Transactions on Graphics 2015;34:135:1-135:12. doi:10.1145/2766937. [22] ROS-Industrial. ROS-Industrial n.d. https://rosindustrial.org/ (accessed March 20, 2019). [23] UR5 collaborative robot arm | flexible and lightweight robot arm n.d. https://www.universal-robots.com/products/ur5-robot/ (accessed March 21, 2019). [24] Kebria PM, Al-wais S, Abdi H, Nahavandi S. Kinematic and dynamic modelling of UR5 manipulator. 2016 IEEE International Conference on Systems, Man, and Cybernetics (SMC), 2016, p. 004229– 34. doi:10.1109/SMC.2016.7844896. [25] Spong MW, Hutchinson S, Vidyasagar M. Robot modeling and control. Hoboken, NJ: John Wiley & Sons; 2006. [26] LASA Laboratory, EPFL. GitHub n.d. https://github.com/epfl-lasa (accessed March 21, 2019).
ScienceDirect ScienceDirect Available online atonline www.sciencedirect.com Procedia CIRP 00 (2019) 000–000 Available at www.sciencedirect.com
ScienceDirect ScienceDirect
Procedia CIRP 00 (2019) 000–000
www.elsevier.com/locate/procedia www.elsevier.com/locate/procedia
Procedia CIRP 00 (2017) 000–000 Procedia CIRP 81 (2019) 713–718
52nd CIRP Conference on Manufacturing Systems 52nd CIRP Conference on Manufacturing Systems
www.elsevier.com/locate/procedia
Hierarchical motion control for real time simulation of industrial robots 28th CIRP Design for Conference, May simulation 2018, Nantes, France Hierarchical motion control realatime of industrial robots a Tadele Belay Tuli *, Martin Manns a a Tadele Belay *,9-11, Martin Manns Siegen,Paul-Bonatz-Str. 57076 Siegen, toUniversität analyze theTuli functional andGermany physical
A new methodology architecture of Siegen,Paul-Bonatz-Str. 9-11, 57076 Siegen, Germany * Corresponding author. Tel.: +49-271-740-2267; fax: +49-271-740-2542. E-mail address: [email protected] existing products for Universität an assembly oriented product family identification a a
* Corresponding author. Tel.: +49-271-740-2267; fax: +49-271-740-2542. E-mail address: [email protected]
Abstract Abstract
Paul Stief *, Jean-Yves Dantan, Alain Etienne, Ali Siadat
École Nationale Supérieure d’Arts et Métiers, Arts et Métiers ParisTech, LCFC EA 4495, 4 Rue Augustin Fresnel, Metz 57078, France
Multi axis machine tools implement real time motion control algorithms. Objects including difficult to manufacture and on the object surface. Force compliant industrial robots Multi axis machine tools implement real time motion control algorithms. Objects including difficult to manufacture and are used to manipulate deformable objects in real time simulation. In this research, a hierarchical method of motion controlling deformable shapes show variable properties when tools apply a pressure on the object surface. Force compliant industrial robots strategy is presented. The proposed approach is described regarding performance characteristics such as accuracy, repeatability, are used to manipulate deformable objects in real time simulation. In this research, a hierarchical method of motion controlling controllability of the motion segmentation and time dynamics. The result shows that a hierarchical control approach can be strategy is presented. The proposed approach is described regarding performance characteristics such as accuracy, repeatability, Abstract considered as a potential candidate to manipulate deformable objects where force requirements are critical. controllability of the motion segmentation and time dynamics. The result shows that a hierarchical control approach can be Inconsidered today’s business environment, the trend towards more product variety andwhere customization is unbroken. are Duecritical. to this development, the need of as a potential candidate to manipulate deformable objects force requirements *deformable Correspondingshapes author. Tel.: 3 87 37 54 30; E-mailwhen address: [email protected] show+33 variable properties tools apply a pressure
© 2019 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license agile andThe reconfigurable production systems Ltd. emerged to cope with various products and product families. To design and optimize production © 2019 Authors. Published by Elsevier (http://creativecommons.org/licenses/by-nc-nd/3.0/) systems as well as to choose the optimal product matches, product analysis areCC needed. Indeed, most of the known methods aim to © 2019 The Authors. Published by Elsevier Ltd. This is license an open access articlemethods under the BY-NC-ND license This is an open access article under the scientific CC BY-NC-ND (http://creativecommons.org/licenses/by-nc-nd/3.0/) Peer-review under responsibility of the committee of the 52nd CIRP Conference on Manufacturing Systems. analyze a product or one product family on the physical level. Different product families, however, may differ largely in terms of the number and (http://creativecommons.org/licenses/by-nc-nd/3.0/) Peer-review under responsibility of the scientific committee of the 52nd CIRP Conference on Manufacturing Systems. nature of components. This fact impedes an efficient comparison and choice of appropriate product familySystems. combinations for the production Peer-review under responsibility of the scientific committee of the 52nd CIRP Conference on Manufacturing Keywords: motion planning; industrial robot; realtime simulation; hierarchical control system. A new methodology is proposed to analyze existing products in view of their functional and physical architecture. The aim is to cluster Keywords: motion planning; industrial robot;product realtimefamilies simulation; control of existing assembly lines and the creation of future reconfigurable these products in new assembly oriented forhierarchical the optimization assembly systems. Based on Datum Flow Chain, the physical structure of the products is analyzed. Functional subassemblies are identified, and a1.functional analysis is performed. Moreover, a hybrid functional and physical (HyFPAG) is the whichsystem depictsand the Background adapt architecture tasks usinggraph learning skills in an output automated similarity between product families by providing design support to both, production system planners and product designers. An illustrative apply multi-level control strategies [6–8]. 1. Background tasks using skills in an automated example of a nail-clipper is used to explain the proposed methodology. Anadapt industrial case studylearning on two product families of steeringsystem columnsand of In multi axes machine tools, objects including difficult to Modern machine tools are capable apply of multi-level control strategies [6–8].to generate tool path thyssenkrupp Presta France is then carried out to give a first industrial evaluation the proposed approach. and machine deformable shapes manipulated using points for machine free forms (e.g. [9], tool complex ©manufacture 2017 The Authors. Publishedtools, by Elsevier B.V. In multi axes objectsare including difficult to Modern tools are turbine capable blades) to generate path real time motion control. Industrial robots as intelligent geometries (e.g. additively manufactured soft-grippers) [10], Peer-review under the scientific committee of the 28th CIRP Design 2018.(e.g. turbine blades) [9], complex manufacture andresponsibility deformableof shapes are manipulated using points forConference free forms
manufacturing toolscontrol. are adapted to new geometric variants real time motion Industrial robots as intelligent in automated manufacturing tools are adapted to new geometric variants production system. The wide range of industrial and mechanical properties using sensor models inapplications automated makes robot motion planning and controlling strategies costly production system. The wide range of industrial applications and complex [1]. Recently, robot motion control is being makes robot motion planning and controlling strategies costly 1.simplified, Introduction and robots are sharing a common working space and complex [1]. Recently, robot motion control is being with humanandbeings [2,3].sharing Some aautomated manufacturing simplified, common working space Due to that therobots fast are development in the domain of processes require low payloads use force compliant with human beings [2,3]. Some automated manufacturing communication and an ongoing trend of digitization and robots. Particularly, increase of force use compliant industrial processes thatmanufacturing require low enterprises payloads force compliant digitalization, are facing important robots with an open architecture, has created a common robots. Particularly, increase ofenvironments: force compliant industrial challenges inboth today’s market a continuing platform for research and manufacturing communities to robots with an open architecture, has created a times common tendency towards reduction of[4]. product development and exchange practical use cases Open architectures for robot platform for both research and manufacturing communities to shortened product Infreedom addition, there is an increasing controllers providelifecycles. scientific toarchitectures implement advanced exchange practical use cases [4]. Open for robot demand of customization, being atprocesses the same[5]. time in a global control laws into manufacturing Direct access controllers provide scientific freedom to implement advanced competition with competitors all over the world. This trend, to low level joint systems such as servo systems and joint controlislaws into manufacturing processes [5]. Direct access which inducing the development from macro to micro kinematics invites to optimize motion planning techniques, to low level joint suchlotassizes servodue systems and joint markets, results in systems diminished to augmenting kinematics invites to optimize motion planning techniques, product varieties (high-volume to low-volume production) [1].
and surface (e.g. forming (e.g. robotic incremental forming) [11]. geometries additively manufactured soft-grippers) [10], However, soft and deformable objects such as additively and surface forming (e.g. robotic incremental forming) [11]. manufactured are difficult to manipulate due to the However, softelastomers and deformable objects such as additively unknown behavior of the material deformation when pressure manufactured elastomers are difficult to manipulate due to the is applied [12,13]. Force sensor model approaches are unknown behavior of the material deformation when pressure of the product range and characteristics manufactured and/or presented by various groups to control motions of industrial is appliedin [12,13]. Force sensor model approaches assembled this system. In this context, the main challengeare in robots to manipulate such objects [14,15].motions presented by various groups to control of industrial modelling and analysis is now not only to cope with single To implement motion algorithms, motion features robots to manipulate such control objects [14,15]. products, a limited product range or existing product families, such as execution time, position accuracy, control strategy, To implement motion control algorithms, motion to features but also to be able to analyze and to compare products define collision avoidance and error interpretations are mainly such as execution time, position accuracy, control strategy, new product families. It can be observed that classical existing considered. Motion control is used to prescribe are a robot to collisionfamilies avoidance and error interpretations mainly product are regrouped in within function of defined clients ortime features. follow a desired trajectory a by considered. Motionoriented controlproduct is usedfamilies to prescribe a robot to However, assembly are hardly to find. complying to force ortrajectory torque requirements. There are time multiple follow a desired within a defined by On the product family level, products differ mainly in two motion control approaches including intelligent based complying to force or(i)torque requirements. Therevision are multiple main characteristics: the number of components and (ii) the control strategies [16], multi heterogeneous sensor motion control approaches including intelligentelectronical). vision based based type of components (e.g. mechanical, electrical, control strategies [17], reality based control strategies control strategies [16],virtual multi heterogeneous sensor based Classical methodologies considering mainly single products control strategies [17], virtual reality based control strategies or solitary, already existing product families analyze the 2212-8271 © 2019 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license To cope with this augmenting variety as well as to be able to product structure on a physical level (components level) which (http://creativecommons.org/licenses/by-nc-nd/3.0/) 2212-8271 © 2019 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license an efficient definition and identify possible optimization potentials in the existing causes regarding Peer-review under responsibility of the scientific committee of the 52nd CIRP Conference on difficulties Manufacturing Systems. (http://creativecommons.org/licenses/by-nc-nd/3.0/) production system, it is important to have a precise knowledge comparison of different product families. Addressing this and mechanical using sensor models Keywords: Assembly;properties Design method; Family identification
Peer-review under responsibility of the scientific committee of the 52nd CIRP Conference on Manufacturing Systems.
2212-8271 © 2019 The Authors. Published by Elsevier Ltd. This is an©open article Published under theby CC BY-NC-ND 2212-8271 2017access The Authors. Elsevier B.V. license (http://creativecommons.org/licenses/by-nc-nd/3.0/) Peer-review under responsibility of scientific the scientific committee theCIRP 52ndDesign CIRPConference Conference2018. on Manufacturing Systems. Peer-review under responsibility of the committee of the of 28th 10.1016/j.procir.2019.03.181
714 2
Tadele Belay Tuli et al. / Procedia CIRP 81 (2019) 713–718 Tadele Belay Tuli and Martin Manns / Procedia CIRP 00 (2019) 000–000
[18], geometrically constrained motion control, and fuzzy based force/position hybrid control [15]. Today, as digital manufacturing concepts have been widely adopted, industrial robots are available with an open controller architecture. Converting ideas into reality is being further simplified. Real time motion planning and control is effectively presented for handling or grasping of rigid objects where object deformation can be negligible. Handling of deformable objects where the amount of applied pressure is critical e.g. hollow micro structures are still open for further investigations. In such cases, a real time adaptable motion generation approach is required. Deformable objects are difficult to handle; first they are highly sensitive to applied force, second a particle of the object changes a position with a small applied pressure. Third, controlling a force at a desired value is difficult [10,13,19–21]. Robot programming system (ROS) is a meta-operating system which provides the desired service of a dedicated robot for hardware in the loop simulation. The operating system includes hardware abstraction, low level control of sensors and actuators, implementation of functionality share, messaging, and package management. In manufacturing automation, robots have some unique functional requirements, job specification, working environment and operation parameters unlike service robots. The initiative of Yaskawa Motoman Robotics, Southwest Research Institute, and Willow Garage, has extended the functionality of ROS into the industrial community with the principal goal of manufacturing automation and robotics. ROS-I runs on the Linux operating system, and it does not provide assurance to robot motion. To overcome this issue, robot designers sub-divided the system into real time and non-real time subsystems. The real time robot motion execution is implemented using embedded system coupling of ROS-Industrial and ROS bridge [22]. The Universal Robot UR5 that are considered in this research is a low payload type industrial robots with six revolute joint variables. It is known to be among the first generation of robots to be used in human robot collaboration. It supports a TCP/IP communication at 100Mbyte [23]. ROS-I provides subscriber and publisher functionality to receive and send controller parameters [4]. 2. Objective In this work, we propose a hierarchical controller approach to control a motion of an industrial robot at real time. The approach considers tool center point force property and joint space variables to derive a robust motion control that hierarchically adapts to the level of object deformation. Additively manufactured elastomer cubes with hollow cross sections are chosen to develop a real time force control. We investigate how to apply pressure on such a deformable object that has a resilient motion property. In this case, a micro motion movement is considered to characterize an applied force. In this regard, we justify that in a real time simulation, acceptable ranges of target force can be achieved by implementing PID controllers into the system. Finally, we verify our approach by implementing a real time simulation using ROS-I. The Python programming language is used to
establish system interface. We develop a function that is used for data acquisition, data analysis and data transmission using ROS-I’s built in APIs . 3. Robot motion modeling Given the initial and final joint space variables such as position, forward robot kinematics computes the global position and orientation of the robot tool center point with respect to a fixed reference frame. Alternatively, inverse kinematics maps the Cartesian space of the robot end effector to the joint space variables. The analytical kinematic model presented in [24] is adopted. Considering the kinematic configuration of six axes of the considered robot, the joint state position vector of wrist three (i.e. frame six) with respect to the base frame (i.e. frame zero) can be given in equation 1. 𝑞𝑞1 (𝑡𝑡) 𝑞𝑞2 (𝑡𝑡) 𝑞𝑞(𝑡𝑡) = 𝑞𝑞3 (𝑡𝑡) ⋮ [ 𝑞𝑞6 (𝑡𝑡) ]
(1)
Where 𝑞𝑞(𝑡𝑡) is the joint state position vector and 𝑞𝑞1 (𝑡𝑡), 𝑞𝑞2 (𝑡𝑡), 𝑞𝑞3 (𝑡𝑡), . . . , 𝑞𝑞6 (𝑡𝑡) are joint state variables of six axis. 3.1. Motion characterization In real time motion control of a robot, desired joint positions are expected at a specific time frame. The joint space variables are broken down into N small segments along the motion axis at a time interval of Ti. The time dynamics of joint variable is set to the minimum actuation time. Considering v, the maximum velocity of the robot and Tmin, the minimum actuation time. The time step used to characterize the motion segment can be computed as; 𝑇𝑇𝑚𝑚𝑚𝑚𝑚𝑚 = min(𝑇𝑇𝑞𝑞𝑞𝑞 )
(2)
where 𝑇𝑇𝑞𝑞𝑞𝑞 is a joint actuation time of joint i. From equation 2, we can express an infinitesimal path segment Δ𝑢𝑢 and actuation time step mathematically as Δ𝑢𝑢 = 𝑣𝑣𝑇𝑇𝑚𝑚𝑚𝑚𝑚𝑚 . In this case, the real time motion generation expects either Tmin or Δ𝑢𝑢 to be constant to maintain a motion that is commanded by velocity. The step size Δ𝑢𝑢 is limited to the discrete servo control interval (i.e. 𝑇𝑇𝑖𝑖 ≥ 𝑇𝑇) , where 𝑇𝑇 is the closure time. Given the joint position 𝑞𝑞𝑖𝑖 (𝑡𝑡), step size (Δ𝑢𝑢𝑖𝑖 ) and step size N can be given as; Δ𝑢𝑢𝑖𝑖 (𝑡𝑡) =
𝑞𝑞𝑖𝑖 (𝑡𝑡) N
(3)
In section 4.3, we characterize joint space variables in order to get smooth, robust and controllable motion profiles by applying equation 3. This approach is important to simplify real time motion generation.
Tadele Belay Tuli et al. / Procedia CIRP 81 (2019) 713–718 Tadele Belay Tuli and Martin Manns / Procedia CIRP 00 (2019) 000–000
715 3
3.2. Motion planning
3.3. Motion Rescaling/adaptation
A trapezoidal velocity profile is a well-known approach for motion planning in the communities of robotics and automated machine tools. However, the ideal trapezoidal velocity profile produces undesired motion behavior at the discontinuities of the velocities. This will impose perturbation to the force controller. To avoid this, we increase the step size (N) and implement a parabolic blending at the discontinuities. Considering N path segments, an intermediate discrete segment k can be modified using time derivative of joint space variables. [16] presented a mathematical procedures to be followed to remove jumps and discontinuities using differential equations. At time instant tk, the time differential of joint space qk(t) possesses a parabolic profile around tk.
In equation 6, there is a piecewise function that interpolates time to ramp up, move constant and ramp down. In a real time implementation of the algorithm, maintaining a constant time for each motion segment is difficult. Instead, we apply a constant segmentation throughout the initial and final position. The discrete approach that maps the tool center position and joint variable at infinitesimal interval is given in equation 7 – 9 by coupling to the robot kinematics (see section 4).
𝑞𝑞̇ 𝑘𝑘−1,𝑘𝑘 = 𝑞𝑞̈ 𝑘𝑘−1,𝑘𝑘 =
𝑞𝑞𝑘𝑘 −𝑞𝑞𝑘𝑘−1
(4)
𝑞𝑞̇ 𝑘𝑘 −𝑞𝑞̇ 𝑘𝑘−1
(5)
Δ𝑡𝑡𝑘𝑘−1 Δ𝑡𝑡𝑘𝑘−1
Where, Δ𝑡𝑡𝑘𝑘 = 𝑡𝑡𝑘𝑘 − 𝑡𝑡𝑘𝑘−1 is the corresponding time of the distance 𝑞𝑞𝑘𝑘 and 𝑞𝑞𝑘𝑘−1 . Accordingly, Δ𝑡𝑡𝑘𝑘−1 ,𝑘𝑘 denotes the time interval to generate a trajectory of 𝑞𝑞𝑘𝑘 and 𝑞𝑞𝑘𝑘−1 .
𝑁𝑁
Case i: Ramping up (i.e. 𝑘𝑘1 ≤ ) 1
3
1
Δ𝑢𝑢1 = (𝑞𝑞𝑖𝑖 + 𝑎𝑎(𝑘𝑘1 Δ𝑡𝑡)2 ) N
2
Case ii: Constant velocity ( i.e.𝑘𝑘2 > 1
1
𝑁𝑁 3
Δ𝑢𝑢2 = ( (𝑞𝑞𝑓𝑓 + 𝑞𝑞𝑖𝑖 − 𝑣𝑣(𝑁𝑁Δ𝑡𝑡)) + 𝑣𝑣(𝑘𝑘2 Δ𝑡𝑡)) N
2
2
& 𝑘𝑘3 ≤ 𝑁𝑁) 3
2
Case iii: Ramping down ( 𝑘𝑘3 > 𝑁𝑁 𝑠𝑠 & 𝑘𝑘3 ≤ 𝑁𝑁) Δ𝑢𝑢3 =
1
N
1
3
ሺሻ
ሺͺሻ
1
(𝑞𝑞𝑓𝑓 − 𝑎𝑎(𝑁𝑁Δ𝑡𝑡)2 + 𝑎𝑎(𝑁𝑁Δ𝑡𝑡)(𝑘𝑘3 Δ𝑡𝑡) − (𝑘𝑘3 Δ𝑡𝑡)2 )ሺͻሻ 2
2
The effectiveness of equation 7 – 9 depends on the selection of sampling size (N) and time step (Δ𝑡𝑡). 4. Controller design Fig. 1. A trapezoidal velocity profile with a parabolic blend.
Implementing a linear segment parabolic blend inside a hierarchical control specifically, in which position and force are dependent on sensor models, it is important to observe the tool center mounted force behavior and characterize it. The algorithm is capable to reduce unwanted motion behaviors such as jerk and discontinuities by creating smooth motion transition. In this case, velocity is characterized in three levels. Initially there is “ramping up” for one third of the closure step time followed by a constant motion for the same time before it is “ramping down” to the desired position. The corresponding notation is summarized in equation 6 (see [19]).
𝑞𝑞(𝑡𝑡) =
1
{
2
1
𝑞𝑞𝑜𝑜 + 𝑎𝑎𝑡𝑡 2 , 𝑡𝑡 ∈ [0, 𝑡𝑡𝑏𝑏 ] 2
(𝑞𝑞𝑓𝑓 + 𝑞𝑞𝑜𝑜 − 𝑣𝑣𝑡𝑡𝑓𝑓 ) + 𝑣𝑣𝑣𝑣 , 𝑡𝑡 ∈ (𝑡𝑡𝑏𝑏 , 𝑡𝑡𝑠𝑠 ] 1
1
𝑞𝑞𝑓𝑓 − 𝑎𝑎𝑡𝑡𝑓𝑓2 + 𝑎𝑎𝑡𝑡𝑓𝑓 𝑡𝑡 − 𝑡𝑡 2 , 𝑡𝑡 ∈ (𝑡𝑡𝑠𝑠 , 𝑡𝑡𝑓𝑓 ] 2
2
ሺሻ
where 𝑣𝑣 , 𝑎𝑎 , 𝑡𝑡𝑎𝑎 and 𝑡𝑡𝑓𝑓 are the mean velocity, acceleration, initial blending time and final blending time respectively. In the same time, we can represent 𝑡𝑡𝑠𝑠 = 𝑡𝑡𝑓𝑓 − 𝑡𝑡𝑏𝑏 and initial and final joint variables using [𝑞𝑞𝑜𝑜 , 𝑞𝑞𝑓𝑓 ].
Parts that are additively manufactured using soft materials such as elastomers and silicon exhibit differently [10,21]. Particularly, it is difficult to use industrial robots to precisely manipulate such objects. It requires robust system that regulates joint space variables state by analyzing force behavior at the tool center point. In this section, requirements are identified to design a controller as it is stated below. 1. 2. 3.
Target force: The robot tool center point applies a pressure along a desired. Tool center point constraint: The robot tool center point has a discrete control points to limit unnecessary motions. Motion step: The motion has to be smooth. Therefore, the motion is characterized into features such as time, velocity and distance.
In addition, control parameters are defined at the high level system and low level subsystem. Target force is a low level controller input whereas a tool center position is an input at the high level controller. The desired outputs are controlled force and position. In section 4.1 and 4.2, control strategies for force and position control are presented.
Tadele Belay Tuli et al. / Procedia CIRP 81 (2019) 713–718 Tadele Belay Tuli and Martin Manns / Procedia CIRP 00 (2019) 000–000
716 4
4.1. Position based discrete control
•
Position controller is a discrete controller that is implemented to constrain the tool center point motion along a desired axis. The initial and final discrete position of the tool center point is specified to avoid robot collision. The main role of position control is to continuously generate a motion of the robot using a computed time to achieve the target motion. If the target force cannot be achieved in a desired working space, the controller will bypass the role of low level controller to avoid undesired motion or collision (refer Fig. 2).
•
Fig. 2. A hierarchical approach motion control.
PID (Proportional – Integral – Differential) control is a classical closed loop output feedback control that is represented mathematically by equation 10 (see [25]). It is used to prescribe how the control law acts. 𝑡𝑡
𝑘𝑘𝑑𝑑 𝑑𝑑𝑑𝑑(𝑡𝑡) 𝑑𝑑𝑑𝑑
𝑑𝑑𝑑𝑑
The list of materials and devices used in this experiment are: laptop (i.e. DELL precision 5520, 32GB RAM, Core i7 laptop), ATI Net F/T sensor, transducer cable (CAN), Ethernet hub, Ethernet to thunderbolt adaptor, customized end effecter and universal robot. Besides, system supporting tools such as Linux OS (Ubuntu 18.04 - Bionic), ROS (Melodic release), and Python 2.7.2 are configured to the hardware system.
Fig. 3. System configuration.
4.2. Force control
𝑢𝑢(𝑡𝑡) = 𝑘𝑘𝑝𝑝 𝑒𝑒(𝑡𝑡) + 𝑘𝑘𝑖𝑖 ∫0 𝑒𝑒(𝑡𝑡)𝑑𝑑𝑑𝑑 +
•
Workspace noises and disturbances are either not measured or not decoupled from the system. The errors exhibited due to incompatibility of the robot controller and the remote controller system are neglected. System isolation of independent processes is not considered.
ሺͳͲሻ
where kp, ki and kd represent the associated gains to the proportional, integral and derivative components respectively.
As Fig. 4 illustrates, the ROS-I system handles TCP/IP connections through the intermediate hub to communicate to robot joint sensors and external force sensor. In ROS-I there are topics that are used to channel joint state, transform information, trajectory action and wrench data through messages [4]. The force sensor mounted at tool center point uses only a subscriber function to collect force data whereas the communication between robot and ROS server is two directional (i.e. subscriber and publisher).
4.3. Software configuration ROS-I (Robot Operating Software Industrial) is a popular open source platform that can be used to implement feedback control strategies including multi input and multi output systems. However, industrial robots that are supported by ROS-I have to possess an open controller architecture. In ROS-I system, two packages are created to interface robot system and force sensor system. Both packages use a launcher that prompts for IP address to establish a connection. The specific topics will be instantiated through the corresponding node to handle messages. During the interface configuration, existing open source codes for universal robot [22] and force sensor [26] are used as a starting point. 5. Experimental set-up and real time implementation 5.1. Experimental set-up The experimental set up is implemented in our lab (see fig. 3). During the experiment:
Fig. 4. Hardware configuration and simulation for the tests.
5.2. Real time simulation For a real time simulation, a set of rules are abstracted. In these rules, static IP address is assigned to a robot and a force transducer in the same network (see Fig. 3). In ROS-I both systems are launched separately to enable the communication of the ROS-I system and hardware systems. A subscriber function is developed to extract position and force data into the main function to analyze the data. A publisher function sends a decision of the controller to the robot.
Tadele Belay Tuli et al. / Procedia CIRP 81 (2019) 713–718 Tadele Belay Tuli and Martin Manns / Procedia CIRP 00 (2019) 000–000
Deformable object that exhibits variable mechanical property when a force is applied is considered to analyze applicability of the proposed approach. Such use case is common in most industrial applications e.g. a force control for foam, cloth, rubber and other material. As it is shown in Fig. 5, target force is defined to apply a pressure on a deformable cube (i.e. additively printed silicon with a dimension of 20x20x20mm3). The applicability of hierarchical controller is analysed at a minimum motion segment and due to resilience of the object during force regulation, adaptive motion planning is proposed.
a.
717 5
holder and shop floor disturbance. The experiment is deliberately conducted in a shop floor (see Fig. 4) in which multiple machines are running (i.e. KUKA Jet, SCARA Bosch, and KUKA LWR). This will lead us to generalize that, the experimental setup can be replicated in an industrial environment. For real time simulation purpose, a force control is configured with an optimal controller gain parameters. The gains are sensitive to system noises. We observed that, for uncontrolled system variables, it is necessary to tune the PID gains to obtain better performance. In our system, we employed kp = 100, kd = 0.1 and ki =0 after optimization of the system.
b.
Fig. 5. Use case definition and workspace setup (F denotes the tool center force measured by external force sensor, and Δx(t) represents the time variable of object deformation).
5.3. Description of result Simulation results are described using performance characteristics such as accuracy, repeatability, motion smoothness, controllability of the motion segmentation and time dynamics. These criteria are used to relate the performance of robot movement, task accomplishment and difficulty level of the job with regard to the experimental setup that is discussed in section 5.1 and 5.2.
a.
b. Fig. 6. A force and time output for a target force of 30N using a hierarchical control law (a) Overall force behaviour (b) A cubic curve fitting for target preservation phase.
During a hardware in the loop simulation, noise is observed from the system configuration. The effect is shown by spikes particularly for a higher applied pressure. In our case, at 30N the robot motion is not smooth as the seeking phase (see Fig. 6). The source of this noise can be associated to robot joint stiffness, transfer system drive, work piece
Fig. 7. Residual error distribution of a PID controller output for gain parameters of (kp = 100, kd = 0.1 and ki =0)
6. Discussion In Fig. 6 – 7 the force characteristic is shown. There are three phases in the experiment. These are approach, seeking and target preservation. In each phase, there exist different time dynamics and motion speeds. In the first phase, the robot moves from its home position to the starting position of force measurement. In this case, the robot moves with 75% of the robot maximum speed. The second phase is characterized as force seeking motion. In this stage the robot force applies a pressure to seek a user defined target force. In this situation, the robot speed is analytically determined based on the motion planning algorithm. If the robot achieves the target force, the force controller will regulate the joint sensor to maintain the desired force. In our use case, the object is deformable and it does not maintain the measured force for some duration of time. Hence, it is necessary to apply a change in position to update the new force value to the target one. The simulation result shows, the force is controlled with an error of ± 0.5N. As it is explicitly shown in Fig. 6(b) spikes are observed in a random distribution. In general, the controller is capable to regulate the applied pressure a maximum of ±1N. An observed peak beyond ±0.5N is 6.67%. If we apply a constant pressure at a fixed robot joint position, the force value decreases gradually. This requires a controller action to update the force to a target force. In this research, potential challenges such as deformable object dynamics control and force control in a case of super slow motion is identified for future work. However, hierarchical control method can be considered as an initial approach to control the sensibility of applied pressure a case of deformable object manipulation for industrial applications.
718 6
Tadele Belay Tuli et al. / Procedia CIRP 81 (2019) 713–718 Tadele Belay Tuli and Martin Manns / Procedia CIRP 00 (2019) 000–000
7. Conclusion Real time motion control algorithms are being used for multi axes machine tools. Objects including difficult to manufacture and deformable shapes have dynamic properties when tools and end effecters apply pressure on the surface. This makes handling of deformable objects often difficult particularly in real time simulation. In this paper, an initial approach that uses a hierarchical motion control strategy is presented. Position control is implemented at a high level to constrain the tool center point motion within a defined working space. In addition, the position controller is proposed to generate joint state trajectories based on trapezoidal velocity profile, which blends the trajectory using parabolic equation. The force controller is a low level controller that regulates the applied force comparing with the target force. The real time simulation is implemented using the ROS-I platform and the Python programming language. Generic use case that satisfies deformability, resilient property and unknown mechanical property is identified and developed using additive manufacturing technology. The result indicates that, even though system disturbances have shown an influence in the smoothness of the data acquired, hierarchical motion controlling strategy has a potential to manipulate deformable objects where an applied tool pressure is critical. The study and comparison of the mechanical properties regarding shape, cross sectional area and mechanical characteristics including the system dynamics are to be considered as future work. References [1] Meyes R, Scheiderer C, Meisen T. Continuous Motion Planning for Industrial Robots based on Direct Sensory Input. Procedia CIRP 2018;72:291–6. doi:10.1016/j.procir.2018.03.067. [2] Mohammed A, Wang L. Brainwaves driven human-robot collaborative assembly. CIRP Annals 2018;67:13–6. doi:10.1016/j.cirp.2018.04.048. [3] Glogowski P, Lemmerz K, Hypki A, Kuhlenkötter B. ROS-Based Robot Simulation in Human-Robot Collaboration. In: Karafillidis A, Weidner R, editors. Developing Support Technologies: Integrating Multiple Perspectives to Create Assistance that People Really Want, Cham: Springer International Publishing; 2018, p. 237–46. doi:10.1007/978-3030-01836-8_23. [4] Koubaa A, editor. Robot Operating System (ROS): The Complete Reference (Volume 2). Springer International Publishing; 2017. [5] Tsardoulias E, Mitkas P. Robotic frameworks, architectures and middleware comparison. ArXiv:171106842 [Cs] 2017. [6] Vonásek V, Penc O, Přeučil L. Guided Motion Planning for Modular Robots. In: Hodicky J, editor. Modelling and Simulation for Autonomous Systems, Springer International Publishing; 2014, p. 217–30. [7] Meyes R, Tercan H, Roggendorf S, Thiele T, Büscher C, Obdenbusch M, et al. Motion Planning for Industrial Robots using Reinforcement Learning. Procedia CIRP 2017;63:107–12.
doi:10.1016/j.procir.2017.03.095. [8] Larsen L, Schuster A, Kim J, Kupke M. Path Planning of Cooperating Industrial Robots Using Evolutionary Algorithms. Procedia Manufacturing 2018;17:286–93. doi:10.1016/j.promfg.2018.10.048. [9] Altintas Y. Manufacturing Automation by Yusuf Altintas. Cambridge Core 2012. doi:10.1017/CBO9780511843723. [10] Manns M, Morales J, Frohn P. Additive manufacturing of silicon based PneuNets as soft robotic actuators. Procedia CIRP 2018;72:328–33. doi:10.1016/j.procir.2018.03.186. [11] Lu H, Liu H, Wang C. Review on strategies for geometric accuracy improvement in incremental sheet forming. Int J Adv Manuf Technol 2019. doi:10.1007/s00170-019-03348-3. [12] Shepherd R. Design of elastomeric composites for the additive manufacturing of robots. Composites at Lake Louise 2017 2017. [13] Otaduy MA, Tamstorf R, Steinemann D, Gross M. Implicit Contact Handling for Deformable Objects. Computer Graphics Forum 2009;28:559–68. doi:10.1111/j.1467-8659.2009.01396.x. [14] Kelmar L, Lumia R. Sensory interactive robot trajectory control using a real-time world model. Journal of Intelligent & Robotic Systems 1992;5:39–47. doi:10.1007/BF00357129. [15] Chaudhary H, Panwar V, Sukavanum N, Prasad R. Fuzzy PD+I based Hybrid force/ position control of an Industrial Robot Manipulator. IFAC Proceedings Volumes 2014;47:429–36. doi:10.3182/20140313-3IN-3024.00062. [16] Corke P. Vision-Based Control. In: Corke P, editor. Robotics, Vision and Control: Fundamental Algorithms In MATLAB® Second, Completely Revised, Extended And Updated Edition, Cham: Springer International Publishing; 2017, p. 537–63. doi:10.1007/978-3-319-544137_15. [17] Mozumdar P, Helal TS, Ababnah A, Akter L. Heterogeneous sensor deployment strategy based on optimal control theory. 2015 International Conference on Electrical Engineering and Information Communication Technology (ICEEICT), 2015, p. 1–6. doi:10.1109/ICEEICT.2015.7307446. [18] Burdea GC. Invited review: the synergy between virtual reality and robotics. IEEE Transactions on Robotics and Automation 1999;15:400–10. doi:10.1109/70.768174. [19] Kaboli M, Yao K, Cheng G. Tactile-based manipulation of deformable objects with dynamic center of mass. 2016 IEEE-RAS 16th International Conference on Humanoid Robots (Humanoids), 2016, p. 752–7. doi:10.1109/HUMANOIDS.2016.7803358. [20] Sanchez J, Corrales J-A, Bouzgarrou B-C, Mezouar Y. Robotic manipulation and sensing of deformable objects in domestic and industrial applications: a survey. The International Journal of Robotics Research 2018;37:688–716. doi:10.1177/0278364918779698. [21] Panetta J, Zhou Q, Malomo L, Pietroni N, Cignoni P, Zorin D. Elastic textures for additive fabrication. ACM Transactions on Graphics 2015;34:135:1-135:12. doi:10.1145/2766937. [22] ROS-Industrial. ROS-Industrial n.d. https://rosindustrial.org/ (accessed March 20, 2019). [23] UR5 collaborative robot arm | flexible and lightweight robot arm n.d. https://www.universal-robots.com/products/ur5-robot/ (accessed March 21, 2019). [24] Kebria PM, Al-wais S, Abdi H, Nahavandi S. Kinematic and dynamic modelling of UR5 manipulator. 2016 IEEE International Conference on Systems, Man, and Cybernetics (SMC), 2016, p. 004229– 34. doi:10.1109/SMC.2016.7844896. [25] Spong MW, Hutchinson S, Vidyasagar M. Robot modeling and control. Hoboken, NJ: John Wiley & Sons; 2006. [26] LASA Laboratory, EPFL. GitHub n.d. https://github.com/epfl-lasa (accessed March 21, 2019).