Self-adaptive control of a two-point contact gripper for the precise handling of compliant objects in industrial robotics

This paper presents a novel adaptive control framework for robotic grippers that handles a wide range of compliant objects by mimicking human grasping behaviour. The proposed system integrates three distinct control strategies: classical Proportional-Integral-Derivative (PID), Proportional-Integral-based Fuzzy Logic Control (PI-FLC), and Reinforcement Learning (RL) to achieve precise and safe force modulation during object manipulation. A two-finger gripper prototype was developed and experimentally validated using objects of varying stiffness levels, including rigid (iron, plastic) and deformable materials (silicone, foam, sponge). Real-time force control was benchmarked against human-defined reference profiles derived from tactile interaction experiments. The results demonstrate that while PID control provides satisfactory performance for rigid objects, it fails to adapt to nonlinear dynamics in soft materials. In contrast, the PI-Fuzzy and RL controllers can achieve superior force tracking, stability, and generalisation, closely aligning with human-like grasping patterns. The PI-Fuzzy controller excels in rule-based adaptability, while RL shows potential in learning optimal strategies across different compliance levels. This study underscores the significance of integrating classical and intelligent control strategies to improve robotic dexterity, safety, and autonomy, particularly in unstructured environments. The findings have meaningful implications for industrial automation, human-robot collaboration, and the effective manipulation of objects with varying stiffness.