A demo to showcase the synchronized sim-real bridge of RohoHive during the deployment of a trained policy . Natural Policy Gradient algorithm from mjrl was used to train the policy directly on hardware to reach diverse targets. This policy can be thought to be a learned Inverse Kinematics controller.
A demo to showcase Robohive's IK and minjerk trajectory generator performing bin shuffling experiments over an extended period of time.
RoboHive's viewer when fired with a real-world robot provides an interactive viewer that is very helpful in prototyping and debugging.
RoboHive's attention to physical realism and the sim2real bridge allows seamless transition of results between sim and real. Depicted on the right is a single policy trained via hsim2real deployed on two D'Kitty robots coordinating to push a box to the target.
Policy trained using TorchRL in simulation and deployed in the real world via RoboHive's sim2real bridge.
RoboHive fully support all robot hardware devices.