CPG-RL: Learning Central Pattern Generators for Quadruped Locomotion

In this letter, we present a method for integrating central pattern generators (CPGs), i.e. systems of coupled oscillators, into the deep reinforcement learning (DRL) framework to produce robust and omnidirectional quadruped locomotion. The agent learns directly modulate intrinsic oscillator setpoints (amplitude frequency) coordinate rhythmic behavior among different oscillators. This approach also allows use DRL explore questions related neuroscience, namely role descending pathways, interoscillator couplings, sensory feedback in gait generation. We train our policies simulation perform sim-to-real transfer Unitree A1 quadruped, where observe disturbances unseen during training, most notably dynamically added 13.75 kg load representing 115% nominal mass. test several observation spaces based on proprioceptive sensing show that is deployable with no domain randomization very little feedback, along states, it possible provide only contact booleans space.