A model-decoupled DAgger pipeline for dual-arm robotic manipulation — human-in-the-loop data collection compatible with arbitrary VLA and reward models.

Real-time collision avoidance and contact-aware dual-arm DLO* manipulation using mink and MuJoCo.

ADAPT is a low-cost, safe dual-arm teleoperation framework for the ADAM robot that enables intuitive collection of Learning from Demonstration datasets through active weight compensation and collision-aware control