Innovative Hybrid-Control Policies Enhance High-Precision Manipulation Under Uncertainty

This study introduces hybrid position-force control policies that improve manipulation tasks by dynamically selecting between force and position control. Researchers present Mode-Aware Training for Contact Handling (MATCH), which adjusts policy action probabilities to reflect mode selection behavior. The effectiveness of MATCH is validated through fragile peg-in-hole tasks, where it outperforms traditional pose-control policies, achieving success rates up to 10% higher and significantly reducing peg break occurrences. In extensive sim-to-real experiments, MATCH demonstrates double the success rate in high noise environments compared to pose policies while applying less force on average, showcasing its efficiency in complex action spaces.