A management of mutual belief for human-robot interaction

Aurélie Clodic, Rachid Alami, Vincent Montreuil

TL;DR

This paper introduces a framework for robots to manage "mutual beliefs" with humans during collaborative tasks. By modeling not just what the robot knows, but also what it believes the human knows (and vice-versa, recursively), robots can better understand and predict human intentions, leading to smoother and safer human-robot interaction.

What Problem Does This Paper Address?

Collaborative tasks between humans and robots require more than just the robot understanding its own goals and the environment. For true collaboration, the robot needs to understand its human partner's perspective – what the human believes, what the human thinks the robot believes, and so on. This "mutual belief" is crucial for predictable and effective interaction, especially in scenarios where direct, unmediated collaboration occurs. Without this, misunderstandings can arise, leading to inefficiencies or even safety concerns, as seen in early human-robot guiding experiments where human partners found the robot's movements unpredictable.

Methods

The core of the paper is a framework for managing "unilateral mutual beliefs" (UMB) from the robot's perspective. This involves defining a hierarchy of beliefs:

  • Robot's beliefs: What the robot knows about itself and the world.
  • Robot's beliefs about human's beliefs: What the robot believes its human partner knows.
  • Robot's beliefs about human's beliefs about robot's beliefs: What the robot believes the human believes the robot knows.
  • And so on, recursively.

To achieve these beliefs, the paper proposes communication schemes. These schemes define how information is exchanged through various communicative acts (verbal, gestures, etc.) and how acknowledgments and confirmations are used to build and verify these beliefs. The framework distinguishes between clear and non-clear communication, allowing the robot to handle ambiguity.

Datasets

Not discussed in the paper. The paper focuses on a theoretical framework and illustrates its application with examples and a described experiment.

Key Insights and Ideas

  • Beyond Self-Knowledge: The critical insight is that for effective collaboration, robots must reason about the shared knowledge and mutual beliefs with humans, not just their own internal state.
  • Recursive Beliefs are Key: The paper highlights the necessity of modeling nested beliefs (e.g., "I believe that you believe that I believe...") to truly understand a human partner's perspective.
  • Communication as Belief Building: Communication is framed not just as information transfer, but as a mechanism for establishing and refining mutual beliefs, with acknowledgments playing a vital role in grounding understanding.
  • Handling Ambiguity: The framework explicitly accounts for non-clear communication, allowing robots to manage situations where human intent is not perfectly understood, which is a realistic challenge in HRI.

Key Results

The paper presents a conceptual framework and communication schemes. It illustrates these with examples, including a "give object" task with the Jido robot and a guiding scenario with the Rackham robot. These examples demonstrate how the proposed belief management and communication strategies can lead to more predictable and understandable human-robot interactions, preventing misunderstandings that arise from a lack of shared knowledge. The paper doesn't present quantitative results but rather a qualitative demonstration of the framework's utility.

Limitations and Future Work

The authors acknowledge that the proposed framework is a step towards modeling human-robot interaction. A key limitation is the practical implementation of these belief hierarchies and the reliance on communication schemes to achieve them. Future work would involve developing robust mechanisms for the robot to infer these beliefs from perception and communication, and to integrate them into its decision-making processes for task execution. The paper also suggests that further research is needed to define conventions for human-robot interaction to make communication even more effective.

Summary updated

Abstract

Human-robot collaborative task achievement requires the robot to reason not only about its current beliefs but also about the ones of its human partner. In this paper, we introduces a framework to manage shared knowledge for a robotic system. In a first part, we define which beliefs should be taken into account ; we then explain a manner to achieve them using communication schemes. Several examples are presented to illustrate the purpose of beliefs management including a real experiment demonstrating a "give object" task between the Jido robotic platform and a human.
BibTeX
@inproceedings{clodic2007management,
  author = {Aurélie Clodic and Rachid Alami and Rachid Alami and Vincent Montreuil},
  title = {{A management of mutual belief for human-robot interaction}},
  booktitle = {HAL},
  year = {2007},
}
Figure