Minimalist Compliance Control

Compliance control is a key technology for safe physical interactions in robotics, typically relying on expensive force/torque sensors. However, the high cost and complexity of these sensors limit their application across many robotic platforms. Recent efforts have explored reinforcement learning as an alternative, but these methods often face sim-to-real gaps, lack safety guarantees, and increase system complexity.

Traditional compliance control methods depend on expensive hardware like force/torque sensors, limiting their application across various robotic platforms. Additionally, while reinforcement learning methods can bypass these hardware constraints, they often face sim-to-real gaps, lack safety guarantees, and increase system complexity.

The study proposes a compliance control method without force sensors, using motor current or voltage signals from modern servos and quasi-direct-drive motors to estimate external wrenches. These estimates are integrated into a task-space admittance controller, achieving stable and responsive compliance control. The method is compatible with various high-level planners and does not require complex learning processes.

• �� Estimate external wrenches using motor current signals, avoiding expensive force sensors.
• �� Integrate estimated wrenches into a task-space admittance controller for stable compliance control.
• �� Applicable to various robot morphologies and compatible with diverse high-level planning strategies.
• �� Calibrate motor parameters to improve estimation accuracy.

Experiments were conducted on various robotic platforms, including a robot arm, a dexterous hand, and humanoid robots. The method's effectiveness was validated across multiple contact-rich tasks such as wiping, drawing, scooping, and in-hand manipulation. Comparisons with traditional baseline methods evaluated position error, orientation error, and the accuracy of contact force estimation.

Experimental results show that the method outperforms existing baselines in position and orientation errors, reducing position error to 15.9±5.1 mm and orientation error to 0.048±0.043 rad. Without force sensors, the external force estimation's mean absolute error is 0.69±0.73 N, demonstrating the method's accuracy.

The method can be directly applied to robotic tasks requiring compliance control, such as assembly tasks in industrial automation and object manipulation tasks in service robotics. Its low cost and efficiency make it widely applicable in these fields.

The method's performance in high-dynamic tasks may be limited by the assumed quasi-static interaction. Additionally, the method requires precise motor parameter calibration, which may affect its generalizability. Future research could optimize automatic calibration of motor parameters to improve performance.