Custom Simulator Trains Robot Policy In Under 2 Minutes

Yacine kache, a research engineer focused on reinforcement learning and robotics, posted about training a control policy for a humanoid robot inside a custom rigid-body simulator called dingsim. The training finished in under two minutes on one RTX 4090 GPU. The same policy transferred directly to a Mujoco environment. The post appears as a reply that also credits @jsuarez for prior teaching.