●builderYou can use this approach to train more robust robotic controllers that know how to recover from execution errors.
●researcherThis method addresses the sparse reward problem in long-horizon robotics by generating high-quality recovery supervision.