Virtual leg compliance improves swing-phase collision response when walking on uneven ground

Author(s): Chang, Henry | Advisor(s): Gravish, Nicholas G. | Abstract: Uneven substrates in natural environments impose a lot of challenges on legged robot locomotion, no matter causing body instability during stance phase or obstacle collisions during swing phase. Through the swing phase, the leg lifts off the ground and arcs forward for searching a secure next foothold. The strategies for swing phase motion control are usually conservative, constantly sensing or observing the environment and re-planning the end-effector trajectory if a collision is detected. Inspired by the fast and stable movement of small insects like cockroaches, our target is to design a passive control strategy that overcomes swing-collision without re-planning the trajectory. We implement a virtual compliance method to swing phase actuation using the direct-drive robotic leg. Through the systematic experiments with different obstacle-height and position, we compare the actuator positional control with virtual compliance control. We find out that positional control mode resulted in the leg swing collision that stuck the leg from moving through the obstacle. However, when the leg is actuated through virtual compliance mode, we observed a successful obstacle negotiation across a range of obstacle-height and position. Also, we seek to improve the accuracy and mobility of the leg virtual compliance control under quasi-static assumptions. Through a gradient-based optimizer, we present an optimal design that lowering the inertia effect of the leg assembly.