Diffusion Policy Based walking Controller for Bipedal Robot (Stoch-Biro) in Unseen environments with latent space observations

Bipedal Robot control in lower level is often a problem which requires the controller, which runs at high frequency. This limits the use of heavy networks like GANs, Diffusion Models and other generative models which runs with high compute. We have designed a Real time Robot controller based on Diffusion Models, which not only powerful enough to capture multiple behaviours with different velocities in single policy, but also generalizes well for unseen environments. Our controller is based on learning with offline Data which is better than online learning in some aspects like scalability, simpler training scheme etc. We have Designed and implemented a diffusion model based policy controller on our custom made Bipedal Robot model named ”Stoch Biro” [1] in simulation. This robot has 6 controllable degrees of freedom with position control. We also demonstrated its generalization capability and high frequency control step generation relative to typical generative models which requires huge on boarding compute.