SAC-Optimized Fuzzy Variable Admittance Control for Lead-Through Teaching of Collaborative Robots

Area of Science:

• Robotics
• Control Systems
• Artificial Intelligence

Background:

• Fixed admittance parameters in collaborative robot lead-through teaching create a trade-off between ease of use and motion stability.
• Existing methods struggle to dynamically adapt control parameters for optimal human-robot interaction during teaching tasks.

Purpose of the Study:

• To develop and validate a novel Soft Actor-Critic optimized fuzzy variable admittance control (SAC-FAC) method.
• To improve the balance between operational ease and motion stability in collaborative robot lead-through teaching.
• To enhance the performance of human-robot collaboration by optimizing control parameters.

Main Methods:

• Implementation of a fuzzy variable admittance controller (FAC) that interprets operator intent from interaction forces and motion.
• Online modulation of the damping coefficient using interpretable fuzzy rules.
• Offline optimization of fuzzy membership functions using Soft Actor-Critic (SAC) in simulation, incorporating a saturation mechanism and advanced training techniques (potential-based reward shaping, performance-gated curriculum learning) to ensure stable convergence and prevent parameter degradation.

Main Results:

• The SAC-FAC method demonstrated significant improvements over hand-tuned FAC on a UR10 robot.
• Reductions in mean trajectory tracking error (19.5%), work per unit path (11.6%), and root-mean-square interaction force (6.8%) were observed.
• Performance benefits were more pronounced on complex compound and 3D ramp trajectories, while maintaining fuzzy rule interpretability.

Conclusions:

• The proposed SAC-FAC method effectively optimizes fuzzy variable admittance control for collaborative robot lead-through teaching.
• SAC-FAC achieves a superior balance between operational ease and motion stability compared to traditional methods.
• The approach offers a robust and interpretable solution for enhancing human-robot interaction in collaborative robotics.