Blind Hexapod Locomotion in Complex Terrain with Gait Adaptation Using Deep Reinforcement Learning and Classification

We present a scalable two-level architecture for Hexapod locomotion through complex terrain without the use of exteroceptive sensors. Our approach assumes that the target complex terrain can be modeled by N discrete terrain distributions which capture individual difficulties of the target terrain. Expert policies (physical locomotion controllers) modeled by Artificial Neural Networks are trained independently in these individual scenarios using Deep Reinforcement Learning. These policies are then autonomously multiplexed during inference using a Recurrent Neural Network terrain classifier conditioned on the state history, giving an adaptive gait appropriate for the current terrain. We perform several tests to assess policy robustness by changing various parameters, such as contact, friction and actuator properties. We also show experiments of goal-based positional control of such a system and a way of selecting several gait criteria during deployment, giving us a complete solution for blind Hexapod locomotion in a practical setting. The Hexapod platform and all our experiments are modeled in the MuJoCo [1] physics simulator. Demonstrations are available in the supplementary video.