Hybrid locomotion of a wheel-legged machine Aarne Halme, Ilkka Leppänen, Sami Salmi, Sami Ylönen Automation Technology Laboratory Helsinki University of Technology PL 5400, 02015 HUT, Finland E-mail firstname.lastname@example.org Tel. +358-9-4513300 Fax. +358-9-4513308 ABSTRACT Legged locomotion is very effective in extreme ground conditions, but bad with high speed. Wheeled locomotion is good in high speed but bad on natural terrain. We introduce a way to combine legged and wheeled locomotion to gain effective natural terrain mobility. The mode of locomotion is called rolking (rolling-walking). In addition to effective mobility rolking mode makes it also possible to measure the shapes and unevenness of the ground only by touching it with the wheel. 1. INTRODUCTION WorkPartner is a futuristic service robot designed for outdoor use. It is presently under development at HUT automation Technology Laboratory as a part of national SMART- machine technology program of TEKES. Mobility is based on a hybrid locomotion system, which combines benefits of both legged and wheeled locomotion to provide at the same time good terrain negotiating capability and large velocity range. Figure 1 illustrates the mobile platform, called Hybtor, on which the manipulation and tooling system is built. Figure 1. Hybtor platform The platform has an active body joint and four legs equipped with wheels. The weight is about 200 kg, including all mechanical components and the components of the energy system, the actuating system and the computing system. The payload is about 40 kg, which is mainly taken by the two-hand manipulator system, which is presently under detail design. The actuation system is fully electrical and the power system a hybrid one with batteries and a 3 kW combustion engine. The locomotion system allows motion by legs only, by legs and wheels powered at the same time or by wheels only. With wheels, the machine can obtain 7 km/hour speed on a hard ground. The WorkPartner project and its advancing in mechatronic design have been reported in two previous CLAWAR conferences (1), (2). The idea of this paper is to continue the series by introducing development which is done for its hybrid locomotion and motion control system. The purpose of the hybrid locomotion system is to provide good rough terrain mobility and a wide speed range for the machine by using the same mechanics. Proper design of both the mechanics and the control system are needed to obtain these properties. Tracks were discarded at the beginning because a mild terrain impact is also very important. In this paper we consider why we ended up to the hybrid locomotion system, how it and its control system is designed and introduce also preliminary test results with the Hybtor platform. 2. STRUCTURE OF THE WHEELED LEG The wheeled leg consists of 3dof mammal type leg and an active rubber wheel. One leg weights, including the wheel, about 21 kg. It is capable to produce about 70 kg continuos and 100 kg peak force upwards in the nominal driving position. The maximum stride length when walking is about 0,7 m. The wheel has two functions: the rounded shape rubber wheel works as a foot in the walking mode and as a wheel in driving mode. When using as a foot the rubber wheel absorbs shocks generated in fast walking. The leg-wheel mechanism has been optimized for use as a hybrid propulsion device. Figure 2. Side view of the legs The actuators of the WorkPartner are identical linear actuators. Each of them consists of a Maxon EC250W 48V electric motor, a gear tailor-made by Rover LTD and a ball screw from SKF (CCBR32x100). The wheel movements are realized by a similar electric motor assembled inside the wheel. The Maxon EC250W motor has a static brake for parking. 3. HYBRID LOCOMOTION Hybrid locomotion means combining the wheeled and legged locomotion modes so that the propulsive force is generated by the wheel and the leg joints simultaneously. Hybrid locomotion could be called also ‘rolking’ (rolling-walking). A term close to this one (roller- walker) has been used previously by Hirose (3), but his rolking robot differs from WorkPartner in that its wheels are not powered. Rolking in this case resembles skiing, but instead of skis wheels are used (however, skis are not active devices like wheels in this case). Rolking works like the following. Consider a normal walking sequence. When a leg is in the supporting state the propulsive force is generating by distributed the moments between the leg joints and the wheel joint. How the distribution is done depends on the terrain properties and the speed of the platform. When the leg is in the transferring phase it is not lifted in the air, but unloaded and moved along the ground by touching it all the time and applying a slight forward moment to the wheel at the same time. All the joints of the leg, hip, thigh, knee and wheel are thus controlled actively all the time. In the transferring phase it is possible to ‘feel’ the shapes of the ground and detect the obstacles by measuring the actuator currents and the joint angles (the leg is in a force control mode). The robot can then move on an uneven terrain by “probing” it like a blind animal. On a highly soft terrain, where wheeled locomotion is difficult or impossible, it has been experimentally observed that rolking motion can improve mobility considerably. Other benefits of the rolking mode compared to normal walking are better speed, stability and weight distribution of the platform. The leg can be moved to supporting phase instantly if needed, which improves the reaction responses. Speed is improved because there is no time wasted when lifting or lowering the leg in the walking cycle. Stability will not be easily lost and the weight distribution is more equally divided because the transferring leg supports itself when moving. Standard gait algorithms can be used. When the gait algorithm command a transferring leg to the supporting phase, it can be done instantly because the leg is already on the ground. This is very effective especially when free gait algorithms are used, which seems to be the natural choice in this case. The only disadvantage of the rolking mode, if compared with normal walking, is that the legs can be moved to the same direction as the wheels are rolling. The motion direction must be thus controlled like in the wheeled mode. In the case of Hybtor steering is done by using the articulated body. On the other hand, changing between the different locomotion modes is very simple and in fact they are all controlled by the same program by only using different parameters. 4. CONTROL OF ROLKING LEG In the rolking mode the leg is controlled periodically over a working cycle which is similar to the corresponding cycle in normal walking including the supporting and transfer phases. In the supporting phase the wheel is under speed control or locked and the leg joints make the propulsive stroke. The joint controllers are designed as combined moment and position controllers. In the transfer phase all joints are controlled in a similar way but with different parameters. The wheel is controlled under moment control which produces a slight moment to the direction of motion. In addition, the force by which the leg supports to the ground is calculated from the currents of the actuator motors. This force is controlled to a given setpoint which is much less than the forces on the supporting legs but enough to keep the leg on ground during the transfer motion. If the leg hits to an obstacle or stability of the machine is to be loosed a logic reflex algorithm takes over and controls the setpoints of the basic controllers. Both the basic control algorithm and the reflex-type algorithm and some results of one leg rolking over an obstacle are described shortly in the following. 4.1 Basic control algorithm In the rolking mode, the wheel is driven with constant speed by the velocity controller. In the same time the ankle point of the leg is force controlled with constant forces 85 N forward and 130 N downward. The movement of the ankle point is observed constantly and if the ankle point moves less than 2 cm forward during 700 ms, it is concluded that there is an obstacle. After detecting the obstacle the leg will be lighten so that in the ankle point there is still 85 N force forward but 30 N upwards. Because rotation of the wheel, it is easier to overrun the obstacle. The lightening of the leg takes 700 ms time. Then the algorithm will be continued from the beginning. Waiting and lightening times and force parameters can be changed according to operation situation. This algorithm works with minimal sensor information, it needs only joint angles and rotation of the wheel. 4.2 Reflex-type algorithms In the rolking mode the transfer leg is first moved towards the new support point, but if there is no ground contact or the ground is slippery or simply does not support the weight of the machine, a reflex-like actions are needed. The reflex like action under studying is the following: when no contact the leg is moved backwards the distance of the wheel radius, and the support force is tested again. This procedure is continued until sufficient contact is found. This is done by the leg control software, but the reflex action is informed to the main computer. When the sufficient contact is found, the leg is switched to support mode and it starts to wait for new commands. The contact is tested by setting a downward force to the ankle of the leg and measuring if the ground can hold it and the leg stops moving downward. The basic algorithms of reflex actions are similar in walking and rolking mode. In the walking mode the leg is lifted, but in rolking mode it is moved backwards the same way as in rolking forward. Rolking is effective also in reflex actions because no time is lost in lifting ands lowering the leg. 4.3 Tests of the overrun of the obstacle In the rolking test of one leg the wheel hits an obstacle, which is higher than the radius of the wheel. The test set up is illustrated in the Figure 3. 260 110 140 Figure 3. The real obstacle In the rolking phase the leg supports lightly the body and the wheel rotates actively by the velocity control. When the wheel hits the obstacle and is not able to go forward, the control algorithm starts to lighten the leg. This happens in the Figures 4 and 5 at 0.7 s. Thigh and knee currents change according to the force reference. Wheel current changes according to the velocity reference. The overrun of the obstacle is based on lightening of the leg and active rolling of the wheel. When the wheel is on the obstacle, lightening of the leg is no more needed and the wheel rotates forward quite freely, this happens at 2.0 s in the Figures 4 and 5. When the wheel is driving down to the lower step under velocity controller, it brakes using negative moment as can be seen in the Figure 4 "Current of the wheel motor". This prevents the wheel fall down suddenly and the wheel overcome obstacle smoothly. The benefit of the leg-wheel structure is that it is possible to find out form of the obstacle. In the Figure 5 ankle point is moving in the Cartesian coordinates. From the graph "Form of the obstacle" the side profile of the obstacle can be outlined. It must be taken into account that the ankle point is radius of the wheel away from the obstacle, this courses rounding of the corners. Current of the thigh actuator -0.5 thigh current [A] -1 -1.5 -2 -2.5 -3 0 0.5 1 1.5 2 2.5 3 3.5 4 time [s] Current of the knee actuator 1 knee current [A] 0 -1 -2 -3 0 0.5 1 1.5 2 2.5 3 3.5 4 time [s] Current of the wheel motor 6 wheel current [A] 4 2 0 -2 -4 0 0.5 1 1.5 2 2.5 3 3.5 4 time [s] Figure 4. Currents of the actuators Movement in the vertical direction -250 x-coordinate [mm] -300 -350 -400 -450 -500 0 0.5 1 1.5 2 2.5 3 3.5 4 time [s] Movement in the horizontal direction 600 z-coordinate [mm] 400 200 0 0 0.5 1 1.5 2 2.5 3 3.5 4 time [s] Form of the obstacle -250 x-coordinate [mm] -300 -350 -400 -450 -500 0 100 200 300 400 500 600 z-coordinate [mm] Figure 5. Cartesian coordinates of the ankle 5. HIGHER LEVEL CONTROL OF HYBRID MOTION In the rolking motion the main functions of the overall control system are almost same as in classical walking. Some new features are, however, needed. The rolking mode could be understood as walking without lifting the legs, but unloading and driving them in the transfer phase. This means that the same type of control strategies as applied in walking can be used in the rolking motion. Gating algorithm, like wave gate or free gate, can be copied from classical walking. In traditional walking algorithms first the leg which can be lifted to transfer phase is chosen. In rolking mode this part of the algorithm is alike. Next the new supporting position where the leg is to be transferred is calculated according to the speed and direction of the machine and the form of the ground. This is also the same in rolking mode. In the walking algorithms next the transfer path is planned, the height and shape of the path and the speed of the leg. This part differs the most in rolking mode. In rolking mode the shape and the height of the transfer path varies with the ground unevenness and can be measured online. The speed and supporting force are set beforehand and the leg controller “lightens” actively the leg so that it moves easily along the ground. 6. CONCLUSIONS The rolking (rolling walking) locomotion mode seems to be a natural and effective way of moving for hybrid wheeled-legged machines. It combines the good features of both legs and wheels in difficult terrain conditions. Terrain unevenness can be detected and relative big obstacles can be negotiated easily. In addition, stability of the machine is maximized because reflexive leg motion to support the body can be easily realized in any moment during the leg transfer state. Experimental tests this far are, however, only preliminary, so final conclusions cannot be made. Exploitation of all the features of the rolking locomotion need still work which is on-going. An interesting and challenging problem is also automatic change between the different locomotion modes. REFERENCES 1. Leppänen I., Salmi S. and Halme A., WorkPartner – HUT-Automations new hybrid walking machine, CLAWAR'98, Brussels 1998. 2. Halme A., Leppänen I. and Salmi S., Development of WorkPartner-robot – design of actuating and motion control system, CLAWAR'99, Portsmouth 1999. 3. Endo G. and Hirose S., Study on Roller-Walker (Multi-mode Steering Control and Self-contained Locomotion), International Conference on Robotics & Automation, San Francisco 2000. The research reported in this paper has been supported by TEKES under the contract no. 40262/99 and by the Academy of Finland by the contract no 40656.