Model Predictive Trajectory Planning for Human-Robot Handovers

TL;DR

The paper tackles reliable human-robot handovers in dynamic settings by advancing a path-following model predictive controller that uses a path progress variable $\phi$ to track handover progress. It couples a BoundMPC framework with Gaussian process regression to predict the handover location and its uncertainty, projecting these predictions onto the path and integrating them via adaptive error bounds and a terminal cost. Key contributions include allowing backward motion along the path, uncertainty-aware bound adaptation, orientation-aware projection, and a synchronization mechanism that aligns human and robot progress, demonstrated on a 7-DoF manipulator. The work enables convergence, safety, and predictable, natural interaction in dynamic handover scenarios, with practical implications for real-time HRI in uncertain environments.

Abstract

This work develops a novel trajectory planner for human-robot handovers. The handover requirements can naturally be handled by a path-following-based model predictive controller, where the path progress serves as a progress measure of the handover. Moreover, the deviations from the path are used to follow human motion by adapting the path deviation bounds with a handover location prediction. A Gaussian process regression model, which is trained on known handover trajectories, is employed for this prediction. Experiments with a collaborative 7-DoF robotic manipulator show the effectiveness and versatility of the proposed approach.

Figures (6)

• Figure 1: Schematic working principle of BoundMPC oelerichBoundMPCCartesianTrajectory2024.
• Figure 2: Visualization of the considered human-robot handover situation. The predicted handover location distribution $\boldsymbol{p} _{\mathrm{HO}}$ which determines the error bounds is visualized by the green circle. As the human is already close the handover location, the adapted handover location $\tilde{ \boldsymbol{p} }_{\mathrm{HO}}$ is a linear interpolation between the current human position $\boldsymbol{p} _{\mathrm{h}}$ and the predicted position $\boldsymbol{\mu} _{\mathrm{HO}}$. The reference orientations at $\tilde{ \boldsymbol{p} }_{\mathrm{HO, 0}}$, $\boldsymbol{p} _{\mathrm{r, a}}$ and $\boldsymbol{p} _{\mathrm{r, 0}}$ are indicated by a gripper symbol.
• Figure 3: Error bounding functions for $e_{\mathrm{p, r}, 1}^{\bot}$ in two consecutive time steps where $\Psi_{\mathrm{p, l}, j, \mathrm{prev}}$ and $\Psi_{\mathrm{p, l}, j}$ are the error functions in the first and second time step, respectively. The bounding functions range from the start of the handover segment at $\phi_{0}$ to the handover location at $\tilde{\phi}_{\mathrm{HO}}$. The previous estimation of the handover location path parameter is $\tilde{\phi}_{\mathrm{HO, prev}}$
• Figure 4: Position trajectories of the robot' s end-effector $\boldsymbol{p} _{\mathrm{r}}(t)$ and human hand $\boldsymbol{p} _{\mathrm{h}}(t)$.
• Figure 5: Path parameter of the robot, the human, and the predicted handover location. Additionally, the joint velocity $\dot{q}_2$ of axis 2 is shown with the velocity constraints $\underline{\dot{q}_2}$ and $\overline{\dot{q}_2}$.