SAFERDRILL CONTROL SYSTEM ARCHITECTURE †
G. Pezzuto , J. Gancet , S. Nabulsi* †
D’Appolonia Spa, Space Application Services,* IAI-CSIC
Abstract: The paper explains the recent improvements in the control architecture of a four-legged, two-roped walking robot conceived and designed for consolidation of almost vertical and irregular rocky walls. Saferdrill robot, whose development is funded by EC under Co-operative Research programme, must operate in harsh conditions and remotely controlled in order to make positioning and performing downhill drilling up to 20 m from a remote and safe place. The system architecture used to coordinate the manoeuvrability of the positioning, the drilling process, and the remote operation of the machine are explained. Copyright © 2006 IFAC Keywords: Control system design, industrial robot, robotics, remote control, users interfaces.
1. INTRODUCTION Application of climbing machines to construction industries is a very active area of research (Armada, 2001). Deep drilling is a common practice for slope consolidation of landslide, where 20 m depth holes are prepared using heavy duty machinery. Saferdrill is a quadruped walking and climbing robot whose developed is funded by the EC under Co-operative Research programme with the objective to develop a tele-operated serviced robotic system that will perform safely consolidation and slope monitoring tasks. The whole concept of the robot is based on a mechanical structure, walking on horizontal uneven terrain and climbing on extremely inclined slopes, with on-board the necessary equipment to consolidate and to monitor the slopes. The total weight of the robot is over 2500 Kg. The system is designed to be manoeuvrable and precise so that the operator can make small position changes in order to find the best place to perform the drilling operations. 2. MECHANICAL ARCHITECTURE Saferdrill robot consists of two main subsystems (Figure 1) (Moronti, 2004). The first one is a mobile platform responsible for generating the walking and climbing gaits necessary to move the robot to a specified position on a horizontal terrain (using only the legs) or on a vertical slope (using legs and ropes).
Fig 1. SAFERDRILL mechanical architecture This subsystem is based on a frame supported by 4 legs and two Tirfors , an hydraulically operated hoist able to grasp and move over a fixed steel rope, anchored on the top of the slope. The second subsystem is a drilling unit responsible for performing autonomous deep drilling operations. A robotic arm moves the rods between the drilling unit and the rod buffer where several rods, each one 1.2 m long, are stored. Several rods are screwed together and to the drilling head to drill up to 20 m depth. Rock is broken by a down-hole hammer powered by pressurised air that removes also the materials from the
hole. Once each hole is drilled, the used set of rods is recovered and stored in the buffer to be reused. The deep drilling operation can be done fully autonomously by the onboard system as a set of control macro have been developed to load and unload the drilling rods using the robotic arm. The two subsystems do not work simultaneously: after the positioning of the robot is made, the legs set the system in a stable and static position; then the drilling tasks start. Additional Saferdrill system components are : -
a group of on-board electric motors and pumps to generate hydraulic power used to actuate legs, rope tensioning devices, drilling unit; an on-board control system and a remote human machine interface (HMI) composed by a Tablet Pc, a console with industrial buttons, joysticks and witches, and a Wi-Fi hotspot; a navigation system composed by an on-board stereo camera and a 3D slope map generation software; a geotechnical knowledge based system able to monitor in real-time all drilling parameters and to alert the operator about possible critical conditions.
The robot is based on a mechanical structure, having a total weight of over 2500 kg, and it is designed to overcome obstacles greater than 500 mm. Each leg is able to support a total weight of 1500 Kg and the four legs are able to carry all the equipment on board that which weighs more than 3500 Kg (Fig. 2). Hydraulic powered cylinders have been selected rather than electrical motors to actuate the robot. The main advantages of the hydraulic servo-actuators are the capability to transmit movements to high
3. Legs degrees of freedom (a) Rotational joint; and (b) Prismatic joints.
payloads, and the ability to be controlled with different types of feedbacks like pressure, flow, position and/or force sensors. Finally, because this application does not need high velocity, actuators can move slowly and provide soft movements; this property can be very useful in the gait generation where the synchronisation of the different actuators is critical. The machine has to work both walking on horizontal uneven terrain and climbing on extremely inclined walls. Each of the four legs has 3 DOF, one rotational and two prismatic joints. Each joint is controlled in speed and in position. A regular encoder measures the angle of the leg respect the axis (Fig. 3(a)) in case of rotational joints. A linear encoder is used as position transducer for prismatic joint (Fig. 3(b)). The legs, positioned around the robot frame like square vertexes, have a rotational span of about 100º. Each leg can extend vertically 700 mm and horizontally about 300 mm.
3. GAIT STRATEGIES Walking and climbing strategies for several environments and working conditions have been the subject of previous investigations by many authors (Song, 1989; Armada, 2003). In particular, climbing strategies have been developed in order to be able to climb on slopes lower than 30º (Hirose, 1997; Nagakubo, 1994). Because of safety and practical reasons Saferdrill has to be hold by steel ropes and helped to be pulled up from the top of the mountain to cope with slopes ranging from 30º to almost 90º. It is then necessary to develop climbing strategies that combine gait generation with the coordination of the required pull up power, so that it can be easily controlled by an operator located in a remote place.
Fig. 2. Saferdrill four legged walking and climbing robot.
Proportional valves control the fluid direction and flow from a hydraulic power supply to an actuation devices according to the requirements of the system. In this case, a double-acting cylinder allows the hydraulic force to be
Fig. 4. Joint Movement: Block diagram of the servohydraulic system with position feed-back for each joint. applied in both directions. The velocity and the position of the actuator is monitored to close the control loop (Nabulsi, 2003) (see Fig. 4). The control unit controls a power drive unit that generates a Pulse Width Modulation (PWM) signal for the position of the aperture of flow in the proportional valve on the desired direction of the cylinder (Jelali, 2003). When simultaneous movements of the legs joints are made, the control loop must be closed by a velocity feedback, waiting the computed time that each joint need to reach a calculated position. Predefined gaits are programmed on the control unit. These gaits are enough complex to position and maintain the robot in a position parallel to the surface, flat or a sloped terrain, in order to guaranty always the stability and safety of the machine. The climbing and walking gaits are different: for the walking process a two face discontinuous gait is applied and for the climbing process a one phase discontinuous gait is generated. The control system is designed to generate the coordination of the joints to move the robot from the kinematics. 3.1 Walking Walking is needed to move the robot to the base of the working area. The robot is going to move on uneven terrain with all the heavy machinery on board. The walking strategy needs to allow a great range of positioning without loosing stability. Under this conditions a two-phase discontinuous gaits is being implemented on the robot control system (Gonzalez de Santos, 2003) (Fig. 5). This kind of gait is safe because there are always at least three legs contacting ground while the main structure is still and the only way the main body moves is when all of the feet are on the ground. One of the main problems of the system is that the movements of the robots are not smooth at the beginning and at the end of each gait step due to the high system inertia. However because the hydraulic system works at low speed, the stability is not affected and the static gait calculations can be used.
Fig. 6. Gait generation diagram As the heavy equipment on board can produce instability, the robot must stay horizontal also on uneven terrain. For this reason the robot uses the information from the ground to adapt to the terrain by changing the position of the rotational join. Two types of ground contact (GC) control have been developed and tested. The first one uses an on/off detection mechanism: when the foot touches the ground a mechanism provokes the activation of a proximity sensor and stops leg movement. The second GC control senses the force of the leg touching the terrain by a sensor arrangement made with strain gauges on the feet. This method gives a better distribution of the forces in each of the legs, but implies much more computational power (Fig. 6). 3.2 Climbing While climbing, the robot is held from the top of the mountain by two steel ropes. Each rope is attached to the front side of the robot by a Tirfor tensioning device. Due the joint use of ropes and legs, the classical stability criteria for climbing robots (Nagakubo, 1994) cannot be applied directly. It is not necessary that the vertical of the center of gravity lies inside the projection of the support polygon in order to maintain the stability (Moronti, 2004). Instead it is important for the stability, to get the center of gravity vertical outside the polygon and behind the line connecting the contact points of the feet.
Fig. 7. Climbing gait movements So the proposed gait (Fig. 7) is composed by an undulatory gait and rope tensioning with the characteristic of having at some gait phases only two legs in contact with the ground. This is a static gait and no dynamic considerations are made. 4. CONTROL ELEMENTS
Fig. 5. Walking gait movements (half cycle).
The general control architecture (Fig. 8) is distributed on the robotic platform and on the HMI unit. The end-user can full control the robotic platform from a secure remote location using the wireless connection between the HMI unit and the on-board single control unit. The HMI unit has been designed to be light and easily carried because
Fig. 10. Signal manager architecture. the correct position and attitude while performing drilling and consolidating work.
the end-user should be able to move around the working area. The HMI unit is powered by a group of batteries into a rucksack while the HMI unit is carried over user's shoulder. The user can control all operations using a industrial buttons, switches and joysticks and can read the telemetry, sensor output, video camera and navigation output on the screen of a tablet pc, used also to change the configuration parameters of the HMI. Integrated into the tablet PC where is the Wi-Fi antenna for the wireless connection with the robotic unit.
The main control systems are based on a control card designed by the Control Department of the Institute of Industrial Automatics (Fig. 9). This card implements a Proportional Integral Differential (PID) with position feedback and with Pulse Width Modulation (PWM). The PWM output is transformed into an analog signal of ±10 Volts necessary to control the hydraulic power units. This kind of electronic cards has the advantage that with only one command the system is able to control various actuators simultaneously and with autonomy; this properties reduces significantly computational power. These cards also allow the implementation of digital and analog inputs and outputs as it is needed to control sensors and external devices.
4.1 On-board control unit
4.2 Signal manager
The on-board control unit and the linked Wi-Fi hotspot are powered directly from the ground by a dedicated 220 V connection. The control unit is responsible to start/stop all engines and pumps, to control the hydraulic valves, to read the signal from the sensors and the video signal from the navigation system. Finally, being powered on a dedicated line, the PC controller can shutdown the output lines of the on-board electric panel in case of emergency stop without loosing the wireless connection with the remote HMI unit. The single on-board control unit is composed by a CPU, control and data acquisition electronic cards, electric power supplies and power conversion. QNX 6.2 real-time based operation system is used to make the reading and the data processing of all sensors information and controlling the different actuators in real time.
The signal management processes the rough commands received from the HMI controls board in order to generate smart requests for the robot and processes the robot status in order to provide relevant information to the operator.
Fig. 8. General control architecture.
The use of many sensors and the coordination of all actions in real-time is necessary for generating the right gait and for the control of the system elements. So the control system has to process information from the sensors in order to maintain the system in
The architecture of the signal manager (Fig. 10) is composed by a Serial Interface (SI) for the communication with the industrial controls board, an Ethernet Interface (EI) for communication with the robot, a Data Conversion and Processing (DCP) component which is the heart of the signal manager, and a Graphical User Interface (GUI) which displays monitored data and relevant error messages to the user. The DCP is shared in two parts: the requests processing (from the controls board toward the robot), and the status processing (from the robot to the GUI and the industrial controls board). Both are based on a "conditions and events" scheme. Indeed, requests and status are linked to “triggers”, defined in terms of the following elements: o o
Fig. 9. Control card design.
A set of pre-conditions: when all the preconditions are satisfied, the trigger is enabled; A set of triggering events: if at least one of the events occurs, AND the pre-conditions are satisfied, then the effect is triggered; A single effect: this trigger type is enabled simply by command activation or error message production.
This generic approach provides a convenient way to "connect" conditions and events (position of the controls on the controls board, robot state...) with given effects. Through the GUI, the operator can easily modify the connections between the controls board and the ctual robot commands.
5.1 Head translation
Fig. 11. HMI control board architecture. In the same manner, the status generated by the robot are linked to effects such as illuminating LEDs onto the controls board, or producing error messages. The effect can be easily customised according to the user preferences and needs. This approach makes the signal manager highly flexible regarding future updates or modification of the robot's commands and status, and the controls board. 4.3 HMI control board The HMI control board (Fig. 11) is the interface between the industrial control panel, made of buttons, witches, joysticks, and the signal manager. A single micro with 1kbit EEPROM and 8 kbit Flash memory has been selected and it is connected trough a I4C bus to 7 general purpose port expanders, each of them with 16 channels that can be programmed as input or output lines. 32 channels are used as digital output at 5Vdc 30mA for the LEDs indicating the position of components on the robotic platform or the status of the commands. 80 channels are used as digital input low-side for buttons and witches. Two analogical inputs are directly connected to the micro. These lines are used to read the signal from 2 proportional joysticks controlling the head rotation and head translation of the drilling unit. A fine control of these two parameters is critical during the drilling operation as described below. 5. ANALOGIC DRILLING CONTROL To control the robot both on translation and in drilling more than fourth different single commands and macros can be issues by the user. Here we will describe in detail the control logic behind the manual control of head rotation and head translation because it is critical during drilling operation. No stability problems are observed during these operation because the inertia and the weight of the moved parts is small compared with the overall inertia of the robotic system.
Fig. 12. Head translation speed law.
During the hole drilling process the head translates at the maximum speed allowed by the hammer drilling capabilities, depending on the local condition of the rock and terrain. During the screw (or un-screw) operation of two rod sections, the translation speed is very low and should be carefully tuned with the head rotational speed to facilitate the threads join. In the contrary the operator should be able to sharply accelerate (or decelerate) the head translation in case of borehole occlusion or discontinuities in the mass rock. For these reasons the operator controls this parameter using a proportional joystick. The control board sends to the signal manager a value between 0-255 proportional to the inclination of it (0-30 degrees). This value represents the “acceleration” and it is used by the signal manager to set the rotational speed. On Figure 12 is shown the increase of speed (oil flow in the hydraulic motor) in a case of different joystick bending. When the user releases the joystick, it comes back in the NULL position but the head continues to rotate to the selected velocity. The user can stop the head translation immediately if he bents the joystick in the opposite direction of the motion (case C). The maximum acceleration (corresponding to the maximum bending) is a parameter that the user can set using the “configuration” window on the tablet PC. 5.2 Head rotation The head rotation is not used in symmetric way. Head clockwise rotation (DX) is the used during drilling operation and to screw the drilling rods together. Head contra clockwise rotation (SX) is used only to unscrew the rods. During normal drilling the rotational speed is the highest but, during rod screwing (or unscrewing) operation, it is very low and needs to be set according to the actual translational speed. The control system uses two maximum acceleration values to control the rotation, one for rotation clockwise and one for rotation contra clockwise. Users can change the default value using the GUI “configuration” window on the tablet Pc. The control of this parameter is shown in Fig. 13. Clockwise rotation (DX). Starting from a condition where the lever is in NULL position and the rotational velocity is ZERO, if the user bends UP the joystick, the head starts rotating in the DX direction. As explained before, the rotational speed will increase depending on the lever inclination. If the user releases the lever, it comes back in the NULL position, and the rotation go head at the same speed. When the user moves the lever DOWN, the rotation stops. When the user releases the lever, it comes back in the NULL position, and the rotation is stopped.
6. ACKNOWLEDGEMENTS Saferdrill project is funded by the EC under Contract N°: COOP-CT-2005-016842. The project partnership is as follows: ICOP S.p.a., Space Applications Services (SAS), IMC Zlin a.s., Comacchio SRL, Zannini Roberto, MACLYSA, D’Appolonia S.p.a., University of GenoaPMAR Laboratory, CSIC-IAI.
Fig. 13 Head rotation speed law. Contra Clockwise rotation (SX). Starting from a condition where the lever is in NULL position and the rotational velocity is zero, if the user pulls DOWN the joystick, the head starts rotating in the DX direction. As explained before, the rotational speed will increase depending on the lever inclination. When the user releases the lever, it comes back in the NULL position, and the rotation goes at the same speed. Now as soon the user moves the lever UP, the rotation stops. If the user releases the lever, it comes back in the NULL position. These control logics are implemented at signal management level in the DCP component.
6. FIRST RESULT Laboratory and field tests have been performed in Italy and Spain. Working on extreme harsh conditions with high vibration and dust, these first tests demonstrated the importance of using electronic and electric components IP65 certified or follows MIL requirements. The control system shows that it is possible to reduce the overtime for positioning the working equipment and drilling a hole of about 20% compared with the actual fully manual operation. Most time saving occurs thanks to the automated loading/unloading rods procedure. The field tests demonstrate also that the user can fully control the whole consolidation process from a remote and safe location using Saferdrill robotic system. An easy tuning of the control setting using the GUI interface has been very useful to tune the machine operations.
7. CONCLUSION The paper explains the recent improvement of the control architecture of a four-legged robot intended for the automation of the firming-up of rocky slopes and walls, to grant safeguard of peopled areas, highways, private residences or public sites. The architecture of the developed control system for the gait management and legs-ropes coordination was introduced. The principal components of the control system have been described as well as the control logic for drilling head translation and rotation.
Armada M., Gonzalez de Santos P. (2001). Perspectives of climbing and walking robots for the construction industry. In: CLAWAR - Climbing and Walking Robots. Professional Eng. Publishing, London. Armada M., Gonzalez de Santos P., Jiménez M. A., Prieto M. (2003). Application of CLAWAR machines. In: International Journal of Robotics Research, vol.22, No 3-4, pp. 251-264, March-April. Hirose S., K. Yoneda, and H. Tsukagoshi (1997). Titan vii: quadruped walking and manipulating robot on a steep slope. In Int. Conf. on Robotics and Automation, Albuquerque, NM. Gonzalez de Santos P., M. A. Jimenez,(2003) Generation of Discontinuous Gaits for Quadruped Walking Vehicles, Journal of Robotic Systems. 12(9), pp. 599611,1995 Jelali M. and Kroll A (2003). Hydraulic Servo-systems: Modelling, Identification and Control. Springer, London. Moronti M., Sanguineti M., Zoppi M., and Molfino R., (2004) Roboclimber: proposal for online gait planning. In : 7th International Conference on Climbing and Walking Robots CLAWAR04, Madrid, Spain. Nabulsi S., Armada M., and Gonzales de Santos P (2003). Control architecture for a four-legged hydraulically actuated robot. In: Measurement and control in Robotics – ISMCR 2003, pp. 291-295, Madrid. Nagakubo A., and S. Hirose (1994), Walking and Running of the quadruped Wall-Climbing Robot Proc. IEEE Int. Conf. Rob. Autom. Pp. 1005-1012. San Diego. Song S. M., K.J. Waldron (1989) Machines that walk: the adaptive suspension vehicle, The MIT Press. Zoppi M., S. Sgarbi, R. Molfino, and L. Bruzzone,(2003), Equilibrium analysis for quasi-static, multi-roped walking robots, Climbing and Walking Robots (CLAWAR 2003), Professional Eng. Publishing, London, UK