Robot Assistance Primitives with Force-Field Guidance for Shared-Task Collaboration

Published in Robotics and Computer-Integrated Manufacturing (2025)

Sophokles Ktistakis1*, Lucas Gimeno1, Fatima-Zahra Laftissi2, Alexis Hoss2, Antonio De Donno2, Mirko Meboldt1

1ETH Zurich  |  2Accenture Labs

Paper

Abstract

This paper proposes a novel framework for human-robot collaboration (HRC) that addresses the critical need for robots to effectively collaborate with humans on shared tasks within unstructured and dynamic environments. While prior research focused on safety-related aspects, such as collision avoidance in shared workspaces, the task-oriented aspects of human-robot collaboration remain largely underexplored. To address this gap, our framework introduces Robot Assistance Primitives (RAPs), low-level robot actions that integrate both safety and task-related behaviors, enabling the robot to function as a collaborative “third hand” across physical and contactless interactions. A key component is an extension of impedance control with virtual force fields, unifying task guidance and collision avoidance. We leverage a state-of-the-art visual perception pipeline for real-time 3D scene understanding and an AR-HMD interface for multimodal task programming. We validate feasibility through technical experiments and conduct a user study on collaborative soldering and assembly, demonstrating significant improvements in efficiency and reduced cognitive load.

Methods

Figure 1: System Overview

Our framework integrates three core components: (1) a real-time RGB-D visual perception pipeline for 3D scene understanding and object detection, (2) a multimodal AR interface for intuitive task programming via gaze, gestures, and speech, and (3) an impedance-based control scheme augmented with virtual force fields to unify task guidance and collision avoidance. These elements enable the execution of Robot Assistance Primitives (RAPs), versatile low-level actions that support both physical and contactless collaboration in unstructured environments

Experiments

Experiment Overview

We conducted a user study with 22 participants performing a collaborative soldering and assembly task, comparing our robotic assistance framework to a manual baseline. The experimental group used Robot Assistance Primitives (RAPs) for PCB handling and assembly, while the control group relied on table-mounted fixtures. We evaluated task completion time, cognitive load (NASA TLX), user experience, and system safety across repeated trials to assess efficiency, usability, and collision avoidance.

Results

Net Time per Run
Figure: Net task execution time across three runs for experimental (Robot) and control groups.
NASA TLX Results
Figure: Average NASA TLX scores comparing first and last runs for both groups.

The experimental group improved task completion time by 55.4% after three runs, although the manual control group remained faster overall. NASA TLX analysis showed a statistically significant reduction in physical demand (p = 0.012), along with lower frustration and higher perceived performance for the experimental group after repeated use. Participants generally found the system engaging and helpful despite some technical issues, and safety checks confirmed minimal collisions and consistent distance maintenance during handovers.

Discussion and Conclusion

The proposed RAP framework enables flexible, task-oriented collaboration by integrating impedance control with force-field guidance and real-time 3D perception, improving safety and reducing cognitive and physical effort during shared tasks. While user studies confirmed its effectiveness and positive user experience, challenges such as slower execution, gesture and speech recognition errors, AR HMD fatigue, and occasional robot instability remain. Future work should address these limitations through hybrid control, improved intent inference, and faster, more predictable motions to fully unlock the framework’s potential for practical human-robot collaboration.

BibTeX

      
  @article{KTISTAKIS2025103061,
  author    = {Sophokles Ktistakis and Lucas Gimeno and Fatima-Zahra Laftissi and Alexis Hoss and Antonio {De Donno} and Mirko Meboldt},
  title     = {Robot assistance primitives with force-field guidance for shared task collaboration},
  journal   = {Robotics and Computer-Integrated Manufacturing},
  volume    = {96},
  year      = {2025},
  pages     = {103061},
  issn      = {0736-5845},     
  doi       = {https://doi.org/10.1016/j.rcim.2025.103061},
}

Contact Information

If you have questions, feel free to contact:

  • Sophokles Ktistakis: ktistaks@ethz.ch