Compliant actuation for humanoid robots (FWOKN193)

This project focuses on research towards bipedal locomotion on uneven terrain, one of the unsolved problems in building humanoid robots. Motivation for doing research in the field of legged robots is their potential for higher mobility compared to wheeled robots, especially in a human environment. Since these machines only need a discrete number of isolated footholds, their mobility in unstructured environments is much higher than their wheeled counterparts, which require a more or less continuous path of support. Many strategies have been developed to maintain the stability of a biped so that the robot is able to walk on level ground or climb stairs. Often the ZMP criterion is used as determining factor of dynamic stability. By introducing compliant actuation passive walkers are able to walk in a highly efficient manner because they exploit the natural dynamics of the system. Previously the author developed the biped "Lucy", actuated with pneumatic artificial muscles, combining trajectory tracking techniques with compliance adaptation. On a higher level a lot of attention goes to path planning. Well-known research in this field is performed by for example Kuffner. He developed navigation strategies for humanoids through complex environments while using their full capabilities both for indoor as outdoor environments. Also to manipulate objects by humanoids in complex environments, planners are developed by Yoshida. Problem description Few robots are really able to walk on surfaces were legged robots have a real advantage over wheeled robots: rough terrain and uneven structures. When the most advanced humanoid robot Asimo gives a show the technical requirements are that the floor surface has irregularities of maximum 2mm and the horizontal deviation is maximum 1°. No slippery or springy floors are allowed. For HRP-2 a stabilizer has been developed that can cope with slightly uneven terrain, the surface may have gaps smaller than 20mm and slopes <5%. However nothing has been published about this stabilizer and is considered secret. Terrain maps can be build using stereo vision, which is a richly studied domain. However, methods for 3D reconstruction of surfaces of real environment are computationally very expensive. This is a disadvantage because it has to be performed real-time and together with other processes as motion planning, trajectory generation, stabilizer,... Given a height map of the terrain and a discrete set of possible footstep motions, planners are developed to generate a sequence of footstep locations to reach a given goal state [Planning Biped Navigation Strategies in Complex Environments]. During a research stay of the author in JRL at Tsukuba, Japan strategies were developed strategies to let the robot HRP-2 dynamical step over obstacles. But also here the footholds had to be flat. Typically errors of such stereo vision data are, for the humanoid robot HRP-2, between -15 and 5mm when the robot stands still. Goal and methodology of the project So there is a clear gap between the unevenness the robot can see and the robot can walk over without initially knowledge of the structure. It is this gap the project focus on using compliant actuation in the ankle joint. Goal of this project is that the robot can step on a non-slippery floor with hobbles. The heights of the hobbles may be up to half the intermediate foot lift. When the hobbles are higher a vision system has to intervene to adapt the desired trajectories. Most of the robots are position controlled using electrical drives making the joint stiff. This control method reaches a desired position whatever the external forces are and will reject any disturbances. In case of an unknown surface this is a disadvantage to have a good and firm contact between sole and ground, where on the other hand flexibility is required. In actuator technology two main sources of compliance exist: by active feedback control and using passive mechanical elements. Active feedback control is typically achieved with an electrical motor, gearbox and a force/torque sensor; the compliance is introduced by the controller. Advantage is that such systems are excellent in tracking control. This is the first part of the strategy. The development of such a strategy is important because most of the robots are powered by electrical motors. The disadvantage of actuators with active compliance is the limited bandwidth, so they cannot absorb impacts and they cannot store energy, useful for walking at higher speeds. Compliant actuators have a significant advantage to cope with impact, energy storage and exploiting natural dynamics. The control of such actuators is more difficult, especially in tracking performances; the control will be studied in a second part of the project. The first type of controller will be implemented on a humanoid robot powered by traditional position controlled motors. By a force/torque sensor in the feet active compliance will be introduced in the control architecture. This part of the research project will be performed at the Italian Institute of Technology (IIT), they have the possession of one of the top research humanoids in the world: a CB of Sarcos. The second part of the project will be implemented in a robot with passive compliant actuators. Such robots cannot be bought commercially. For this research an improved version of the biped Lucy will be build. The sagittal robot will have a joint toe for bigger step lengths, more natural and faster walking. The patented MACCEPA actuator, developed by our research group, will be used, because the torque and compliance can be controlled independently, each by a dedicated servo motor. This actuator has the advantage that by taking out the spring, the actuator is position controlled and the first type of controller can be tested. The MACCEPA is build with off-the-shelf components and is easier to control than if pneumatic actuators are used as in the original Lucy. The control architecture will consist of different parts. A trajectory generator calculates dynamic stable trajectories out of objective locomotion parameters (step length, step height, foot lift and speed) using a ZMP stability criterion. A second unit is a trajectory tracking controller to control the position of the hip and knee joints. A third controller will control the torque and compliance at the ankle joints and if necessary the trajectories will be adapted to maintain stability under disturbances. First the strategy will be developed in simulation and afterwards implemented in the real biped. Conclusion The author has experiences in building bipedal walking robots, bipedal locomotion control and compliant actuation so this project is a logical continuation of his research activities in an unsolved area as walking on rough terrain.

Keywords:mechanica

Disciplines:Other engineering and technology