Task Adaptation in Industrial Human-Robot Interaction: Leveraging Riemannian Motion Policies

TL;DR

A motion control framework that removes the need for manual control of the robot's movement, facilitates the formulation and combination of complex tasks, and allows the seamless integration of human intent recognition and robot motion planning is proposed.

Abstract

In real-world industrial environments, modern robots often rely on human operators for crucial decision-making and mission synthesis from individual tasks. Effective and safe collaboration between humans and robots requires systems that can adjust their motion based on human intentions, enabling dynamic task planning and adaptation. Addressing the needs of industrial applications, we propose a motion control framework that (i) removes the need for manual control of the robot's movement; (ii) facilitates the formulation and combination of complex tasks; and (iii) allows the seamless integration of human intent recognition and robot motion planning. For this purpose, we leverage a modular and purely reactive approach for task parametrization and motion generation, embodied by Riemannian Motion Policies. The effectiveness of our method is demonstrated, evaluated, and compared to \remove{state-of-the-art approaches}\add{a representative state-of-the-art approach} in experimental scenarios inspired by realistic industrial Human-Robot Interaction settings.

Figures (7)

• Figure 1: A motivating application for this work is obstacle grasping with unknown task hierarchy. The robot possesses motion policies for handling both objects but initially prioritizes the orange one for pickup. However, the human operator envisions a different order and issues corrective commands via a gamepad to convey their desired direction of motion. The task is dynamically adapted to pick up the blue object instead.
• Figure 2: Motion control framework for dynamic task adaptation: The human operator observes the robot's motion ${\dot{\bm{q}}_{R}}$ and issues corrective commands $\bm{u}_{H}$, initiating the adaptation of the current task and corresponding adjustment of the planned motion in the RMP planner.
• Figure 3: A 2D task adaptation example, resembling a top-down view of the scene in \ref{['fig:teaser']}: The human-desired direction of motion $\bm{\nabla}\Phi_{des}$ aligns better with the first policy $\bm{\nabla}\Phi_{1}$. The task adaptation prioritizes it $\alpha_1\uparrow$ over the second one $\alpha_2\downarrow$.
• Figure 4: Experimental setup: cylindrical surface with inspection targets for the Franka Panda arm, user input via Gamepad, and visual feedback in RViz. The inertial world frame $\mathcal{F}_W$, as well as the instantaneous surface normal $\bm{n}_S$ and tangent $\bm{v}_S$ are indicated.
• Figure 5: Desired sequence of inspection targets and sensor rotations.

Theorems & Definitions (11)

• Remark
• Remark
• Remark
• Remark
• Remark
• Remark
• Remark
• Remark
• Remark
• Remark