E2HiL

New robot learning method achieves 99% success in 30 minutes

Researchers have developed a new training method called Max-Q Selective Imitation for human-in-the-loop online reinforcement learning in robots. This method uses an MC Q-chunk critic to evaluate action values from inter…