Learning Based Adaptive Force Control of Robotic Manipulation Based on Real-Time Object Stiffness Detection

TL;DR

This paper presents a learning-based adaptive force control method for robotic manipulation that uses real-time object stiffness detection to improve stability and efficiency, especially in medical applications.

Contribution

It introduces an adaptive controller with a data-trained Adaption Module that adjusts control parameters based on force feedback and stiffness detection.

Findings

Achieves fast convergence in force exertion tasks.

Demonstrates stable force control across different arm zones.

Parameters are learned entirely from data.

Abstract

Force control is essential for medical robots when touching and contacting the patient's body. To increase the stability and efficiency in force control, an Adaption Module could be used to adjust the parameters for different contact situations. We propose an adaptive controller with an Adaption Module which can produce control parameters based on force feedback and real-time stiffness detection. We develop methods for learning the optimal policies by value iteration and using the data generated from those policies to train the Adaptive Module. We test this controller on different zones of a person's arm. All the parameters used in practice are learned from data. The experiments show that the proposed adaptive controller can exert various target forces on different zones of the arm with fast convergence and good stability.