Design and Control of a Robotic End-Effector for Adaptive Surface Grinding

This thesis presents the design and development of a low-cost robotic end-effector and adaptive control framework for surface grinding tasks in environments lacking precise system models and high-end force-sensing equipment. The project integrates passive mechanical compliance with model-free, adaptive control strategies to regulate contact force during grinding operations. The mechanical subsystem features a spring-actuated compliant structure that mitigates force fluctuations and enhances safety and stability when interacting with workpieces. Complementing this, the control architecture is built on a PID-based force regulation system, incorporating three tuning methodologies: a gradient-based baseline tuner, a Genetic Algorithm (GA), and a Reinforcement Learning (RL) agent. Both GA and RL methods enable robust, model-free adaptability to dynamic changes in the environment without the need for manual retuning. All controllers were first developed in simulation and then transferred successfully to physical hardware via a sim-to-real pipeline. Experiments demonstrate that the GA-tuned controller achieves high stability and accuracy, while the RL-based controller provides strong adaptability under varying task conditions. The resulting system offers a practical, accessible solution for force-sensitive robotic automation, reducing reliance on expensive sensors and expert calibration. It serves as a foundation for future development of intelligent, low-cost robotic tools suitable for small manufacturers, research laboratories, and educational programs.