Electric Wire Connector Assembly Task

  In contemporary production lines, automatic insertion of electircal connectors is executed almost exclusively under pure position control.  Such strategies remain vulnerable to small pose deviations and are unable to guarantee the delicate contact forces required for high‑reliability seating. Impedance control provides a principled way to shape interaction forces, yet its optimal stiffness–damping parameters depend on connector geometry and on the stochastic tilt that arises from manufacturing tolerances and sensor noise. Deriving these parameters online—and enlarging the admissible pose‑error envelope—can be formulated as a reinforcement‑learning (RL) problem. By observing insertion outcomes, an RL agent learns to tune impedance gains and end‑effector motions that minimize residual misalignment while keeping contact forces within specified limits. To accelerate exploration, the policy is trained in a high‑fidelity simulation environment; closing the sim‑to‑real gap through calibrated contact models and domain‑randomization techniques is therefore a central focus. This combined framework yields an assembly controller that sustains robust connector insertion across wide variations in initial error and part type, surpassing the limitations of conventional position‑only methods.

Variable Stiffness Mechanism and Optimal Control

  This research presents the design, development, and control of a novel multi-degree-of-freedom Variable Stiffness mechanism (VSM) for robotic environmental interaction. The proposed VSM addresses the limitations of conventional mechanisms such as size, weight, and limited controllable stiffness through the development of a novel mechanism that enables independent stiffness modulation in both vertical and lateral directions. Starting with a basic prototype to verify the variable stiffness principle, the design evolves into a compact, lightweight, and high-performance system by incorporating a tunable beam structure for precise stiffness adjustment. A switchable design approach is also introduced, allowing the generation of custom spring profiles tailored to specific task requirements. Furthermore, a cascaded impedance optimal control strategy is developed to facilitate smooth transitions between compliant exploration and firm insertion while minimizing interaction forces to prevent part damage. The VSM is validated through theoretical modeling, simulations, and experiments, showing its effectiveness in complex assembly tasks such as multi peg-in-hole insertion and connector assembly.

This work establishes a versatile framework for stiffness adaptation in robotic systems, with potential applications in high-precision electronic component assembly, complex manufacturing processes, and advanced robotic automation.