AI News

PRM4RL: Dense Rewards from LLM Progress Functions for Long-Horizon RL

ABC1 min read
Robotics