Achieving bipedal locomotion on rough terrain through human-inspired control

Abstract

This paper presents a method for achieving robotic walking on rough terrain through Human-Inspired Control. This control methodology uses human data to achieve human like walking in robots by considering outputs that appear to be indicative of walking, and using nonlinear control methods to track a set of functions called Canonical Walking Functions (CWF). While this method has proven successful on a specific well-defined terrain, rough terrain walking is achieved by dynamically changing the CWF that the robot outputs should track at every step. To make the computation more tractable Extended Canonical Walking Functions (ECWF) are used to generate these desired functions instead of CWF. The state of the robot, after every non-stance foot strike, is actively sensed and a new CWF is constructed to ensure Hybrid Zero Dynamics is respected for the next step. Finally, the technique developed is implemented on different terrains in simulation. The same technique is adopted experimentally on the bipedal robot AMBER and tested on sinusoidal terrain. Experimental results show how the walking gait morphs based upon the terrain, thereby justifying the theory applied.