Planning Human-Robot Co-manipulation with Human Motor Control Objectives and Multi-component Reaching Strategies

TL;DR

The paper tackles enabling robust human-robot co-manipulation under goal uncertainty by embedding human motor control principles into online planning. It fuses speed-accuracy and cost-benefit trade-offs with a two-component reaching model to generate human-like trajectories and a Gaussian-process-informed transition for authority handover. The authors formulate a discrete-time trajectory optimization framework with closed-form expectations under Gaussian state uncertainty and validate it on co-manipulation synchronization and authority handover tasks, showing accurate velocity patterns, goal inference, and smooth transitions. The approach achieves more legible, data-efficient collaboration by aligning robot motion with human motor strategies and provides a practical pathway for real-time robotic assistance in uncertain-work environments.

Abstract

For successful goal-directed human-robot interaction, the robot should adapt to the intentions and actions of the collaborating human. This can be supported by musculoskeletal or data-driven human models, where the former are limited to lower-level functioning such as ergonomics, and the latter have limited generalizability or data efficiency. What is missing, is the inclusion of human motor control models that can provide generalizable human behavior estimates and integrate into robot planning methods. We use well-studied models from human motor control based on the speed-accuracy and cost-benefit trade-offs to plan collaborative robot motions. In these models, the human trajectory minimizes an objective function, a formulation we adapt to numerical trajectory optimization. This can then be extended with constraints and new variables to realize collaborative motion planning and goal estimation. We deploy this model, as well as a multi-component movement strategy, in physical collaboration with uncertain goal-reaching and synchronized motion tasks, showing the ability of the approach to produce human-like trajectories over a range of conditions.

Figures (9)

• Figure 1: Proposed human motor control models (blue), combined with task parameters (purple) into a trajectory optimization problem to produce the robot trajectory. This base model can be extended with observed motion or transitions between ballistic and corrective motion to build collaborative scenarios.
• Figure 2: Multi-component strategy model for transition point prediction using Gaussian Process Regression trained on the experimental data from peternel2019target. The transition position expressed as is the distance from the goal and depends on the goal size and distance. The goal distance from the initial point is normalized to the arm length.
• Figure 3: The top two graphs show the velocity profiles of reaching movements at different target widths (which changes the task difficulty) and distances resulting from the proposed approach. The matching vertical lines show the predicted finish time from Fitts' law. A direct comparison of finishing times from the proposed model and Fitts' law is shown on the bottom using parameters as described in Section \ref{['sec:validation']}.
• Figure 4: Dispersion of trajectories after repetitive reaching from initial position at $[0, 0]$m to the goal at $[0.45, 0]$m (top). Distribution of final $y$ values at the goals with two different widths (bottom), where the blue bars represent the number of hits at different locations, while the orange curves are fitted Gaussians with covariance $2.8e^{-2}$ and $5.4e^{-2}$ for goal width 0.02m and 0.04m, respectively. The dispersion develops along the trajectory, and final dispersion distribution matches well to results in human motor control.
• Figure 5: Position/velocity phase plots from different goal distances (top) and different initial velocities (bottom). It can be seen that the planning problem results in reasonably scaled trajectories over a range of start conditions, and is capable of handling non-zero initial velocities.