Record STMT-W0DF6B1Aspirational

ToP-ToM: Trust-aware Robot Policy with Theory of Mind

Position paper Institutional policyGlobaladopted

This research paper investigates a trust-aware robot policy incorporating a Theory of Mind (ToM) model in a multiagent human-robot collaboration setting. The authors demonstrate that optimizing solely for team performance can lead a robot to utilize reverse psychology, which threatens human trust and may cause trust collapse. To mitigate this issue, the proposed ToP-ToM model uses a robot ToM framework to infer human trust beliefs and dynamically adjust the reward function during reinforcement learning. Experimental results in a card game simulation show that ToP-ToM effectively balances human trust maintenance and overall team performance.

Summary and content fields are drawn from the document text and are pending editorial review.

Who issued it

arXivAcademic

Publisher

Legal forceAspirational

Under whose authorityInstitutional policy

Judged fromfull text

Who it governs

Researchers and academicsAI developers and providers

AI and robotics researchers interested in modeling human-robot trust and Theory of Mind in multiagent systems.

Sectors

Research and science · robotics and artificial intelligence research

What the document contains

What's covered

The paper covers the design and evaluation of a trust-aware robot policy using Theory of Mind (ToP-ToM) within a multiagent simulation environment. It models human trust dynamics and behavioral policies using Beta distributions and risk coefficients, and evaluates four different robot decision strategies.

What's missing

The paper does not test the proposed model with actual human participants in physical robotic settings, nor does it provide legal or regulatory governance frameworks.

Key takeaways

Optimizing robot policies strictly for team performance can cause robots to employ reverse psychology, risking human trust collapse.
The proposed ToP-ToM model integrates human trust beliefs, including true and false beliefs, into the robot's decision-making process.
A dynamic trust-aware reward function helps balance trust maintenance and team performance during human-robot collaboration.
Offline reinforcement learning using Conservative Q-Learning (CQL) effectively trains trust-aware robot decision models from static interaction data.

Outline

· I. Introduction
· II. Related Works
· III. Methodology
· IV. Simulation Results
· V. Conclusions and Discussion

Issued7 November 2023

Length5,695 words · 7 pages

FormatNot recorded

Scopeglobal

Physical AI Governancecontext

robotics theory_of_mind trust_aware_ai reinforcement_learning human_robot_interaction