TAM: Torque Adaptation Module for Robust Motion Transfer in Manipulation
Researchers have developed a Torque Adaptation Module (TAM) to improve robot motion transfer across different hardware and payloads. TAM learns to adjust torque commands, enabling policies trained in simulation to perform robustly on real robots without requiring extensive retraining or domain randomization. This module has demonstrated success in zero-shot execution on a Franka Panda robot for tasks like box pushing and balancing. AI
IMPACT Enables more reliable deployment of trained robotic policies in real-world scenarios with varying hardware.