Balancing and Static Walking Control for a Compliantly Actuated Quadruped Exploiting System Inherent Elasticities

Abstract

Soft actuators have been designed as a promising novel technology to achieve robust and natural-like robotic behaviors. The present thesis addresses the control of a quadrupedal robot, equipped with series elastic actuators (SEA), to achieve balancing and quasi-static locomotion on rough terrains. Thousands of years of biological evolution proofed static gaits, in combination with the elasticity provided by muscles and tendons, to be the most effective ones on challenging surfaces. The balancing of walking machines has to deal with different discontinuous and highly non-linear contact situations; additionally, SEAs provide a decoupling between actuation and end-effectors. Overcoming the latter with a torque feedback at joints level, which reshapes the springs dynamics, showed robustness issues in real world applications, especially for highly compliant systems, which are designed to efficiently execute highly dynamic locomotion tasks. Therefore, in this context, we present a methodology which exploits directly the inherent spring dynamics by controlling the system equilibrium point. The locomotion task is then fulfilled designing a step planner which, taking advantage of an intuitive set of task coordinates, uses only contact detection and the robot attitude to sense the terrain surface and properly control the load distribution during the walking phases succession. The proposed method is evaluated in simulated environments with and without obstacles, to compare the achieved performances and, finally, tested during experiments on the DLR's quadrupedal robot, Bert, showing a consistent robustness on a variety of ground surfaces.