Force-Based Control for Reduced Deformation Human-Robot Transport of Flexible Objects

Abstract

This thesis considers force-based human-robot collaborative transportation (co-transport) of flexible objects using only force/torque measurements at the robot end-effector, which are standard on many industrial robots, without relying on external sensing in the control loop. The challenge is that, unlike rigid-object co-transport, the dynamics of a flexible object distort the interaction force sensed by the robot. As a result, standard admittance control can lead to reduced gain margins, increased deformation, and instability. Large deformations are undesirable because they degrade transport performance and can damage flexible objects.The thesis develops two model-based methods to address this problem. First, a dynamics-compensated admittance controller improves within-trial co-transport performance by compensating for the flexible-object dynamics. Experimental results show a 45% reduction in RMS deformation compared to standard admittance control while allowing higher effective admittance without loss of stability. Second, for repeated co-transport tasks, a dynamics-based estimator reconstructs human motion from measured interaction force and robot motion, and this estimate is used to iteratively correct the admittance-controller error across trials. Experimental results on a UR10e industrial robot platform in a seven-subject preview-tracking study show a 79% reduction in RMS deformation relative to admittance control alone. In the final iterations, the learned motion reduces deformation below the human RMS reference-tracking error. These results show that modeling flexible-object dynamics can make force-based co-transport practical without external sensing in the control loop, and provide a basis for low-deformation collaborative manipulation on industrial robot platforms.