Ground contact and reaction force sensing for linear policy control of quadruped robot
This work addresses the problem of efficient and robust control for walking robots, which is incremental as it builds on existing linear policy methods by adding specific state variables.
The study tackled the challenge of controlling quadruped robots on uneven terrain by augmenting a linear policy's observation space with ground contact and reaction force data, resulting in policies with improved survivability, stability against disturbances, and adaptability to untrained conditions.
Designing robots capable of traversing uneven terrain and overcoming physical obstacles has been a longstanding challenge in the field of robotics. Walking robots show promise in this regard due to their agility, redundant DOFs and intermittent ground contact of locomoting appendages. However, the complexity of walking robots and their numerous DOFs make controlling them extremely difficult and computation heavy. Linear policies trained with reinforcement learning have been shown to perform adequately to enable quadrupedal walking, while being computationally light weight. The goal of this research is to study the effect of augmentation of observation space of a linear policy with newer state variables on performance of the policy. Since ground contact and reaction forces are the primary means of robot-environment interaction, they are essential state variables on which the linear policy must be informed. Experimental results show that augmenting the observation space with ground contact and reaction force data trains policies with better survivability, better stability against external disturbances and higher adaptability to untrained conditions.