Multi-Task Learning of Active Fault-Tolerant Controller for Leg Failures in Quadruped Robots

Electric quadruped robots used in outdoor exploration are susceptible toleg-related electrical or mechanical failures. Unexpected joint power loss andjoint locking can immediately pose a falling threat. Typically, controllerslack the capability to actively sense the condition of their own joints andtake proactive actions. Maintaining the original motion patterns could lead todisastrous consequences, as the controller may produce irrational output withina short period of time, further creating the risk of serious physical injuries.This paper presents a hierarchical fault-tolerant control scheme employing amulti-task training architecture capable of actively perceiving and overcomingtwo types of leg joint faults. The architecture simultaneously trains threejoint task policies for health, power loss, and locking scenarios in parallel,introducing a symmetric reflection initialization technique to ensure rapid andstable gait skill transformations. Experiments demonstrate that the controlscheme is robust in unexpected scenarios where a single leg experiencesconcurrent joint faults in two joints. Furthermore, the policy retains therobot's planar mobility, enabling rough velocity tracking. Finally, zero-shotSim2Real transfer is achieved on the real-world SOLO8 robot, countering bothelectrical and mechanical failures.