Showcase

I thought I'd share a project I've recently done with Bittle in case anyone is interested in exploring simulation with Bittle in a non reinforcement learning application, although one could also use Mujoco for reinforcement learning also.

I used the URDF created for Bittle here. AIWintermuteAI made the URDF files for Bittle a while back, which was a great starting point to get a model of Bittle into the Mujoco simulator made by google deepmind.

My project aimed to use model predictive control (MPC) to generate optimal gait patterns to run open loop on the Bittle robot. The first step was to get a model into the simulator. To do this, I converted the URDF model to XML format for the simulator. Then I created the environment and configuration to use Mujoco Model Predictive Control (MJPC) with Bittle. MJPC was also developed by google deepmind.

MJPC effectively performs forward rollout…

@rhongzongli - I modified the bittle back with a dremel so it fits on the bot when I do not have the Pi attached. It would be nice to get a back that already has a nice cutout for the connector when ordering a presoldered connector or when someone knows they will be adding one. Maybe if every back printed with outline of a "knock-out" printed on the inside, it would allow those with the connector to precisely cut out the hold with a utility knife?