Leveraging Passive Compliance of Soft Robotics for Physical Human-Robot Collaborative Manipulation

TL;DR

This study benchmarks a large-scale soft robotic co-manipulation platform against human-human performance on translational tasks involving a long object. It combines a passively compliant three-link bellows arm with an omni-directional base, controlled by a displacement-based base controller and a model-reference adaptive controller for the arm, within a VR-enabled experimental protocol. Results show the soft robot achieves comparable qualitative interaction quality and substantial quantitative performance, though humans still outperform in several metrics, highlighting both viability and avenues for improvement in speed, control, and arm consistency. The work establishes a foundational benchmark for future optimization of hybrid soft-mobile manipulators in collaborative manipulation tasks with direct human interaction and safety guarantees.

Abstract

This work represents an initial benchmark of a large-scale soft robot performing physical, collaborative manipulation of a long, extended object with a human partner. The robot consists of a pneumatically-actuated, three-link continuum soft manipulator mounted to an omni-directional mobile base. The system level configuration of the robot and design of the collaborative manipulation (co-manipulation) study are presented. The initial results, both quantitative and qualitative, are directly compared to previous similar human-human co-manipulation studies. These initial results show promise in the ability for large-scale soft robots to perform comparably to human partners acting as non-visual followers in a co-manipulation task. Furthermore, these results challenge traditional soft robot strength limitations and indicate potential for applications requiring strength and adaptability.

Figures (9)

• Figure 1: Physical human-soft-robot co-manipulation with a passively compliant robot arm.
• Figure 2: Top-down, annotated view of the mobile base platform and the pressure system. Annotated components include: (A) wheels and hub motors, (B) 90$^{\circ}$ castor gearbox, (C) castor motor, (D) pressure manifold intake, (E) primary pressure manifold assembly, (F) pressure tank access valve, (G) high pressure storage tank, (H) pressure regulator, and (I) pressurized air transmission line to arm.
• Figure 3: Overview of the co-manipulation object (CMO). The object features an end-effector attachment point (left) and handles for human grasp (right), all mounted to force/torque sensors.
• Figure 4: Visualization of displacement-based control algorithm.
• Figure 5: Overview of the virtual environment setup for the co-manipulation tasks.