Record STMT-W4A7553Aspirational

Reward Shaping for Building Trustworthy Robots in Sequential Human-Robot Interaction

White paper State lawGlobaladopted

This research paper addresses the trade-off in human-robot interaction where maximizing task rewards can reduce human trust, while prioritizing trust can compromise performance. The authors formulate sequential human-robot interaction as a two-player Markov game and use a potential-based reward-shaping technique to improve human trust with a bounded performance loss. Applied to an experience-based trust model, the framework yields a linear program that is evaluated in a simulated search-and-rescue mission.

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

Who issued it

arXivAcademic

Publisher

Legal forceAspirational

Under whose authorityState law

Judged fromfull text

Who it governs

Researchers and academicsAI developers and providersAI deployers and users

Researchers and robotics developers looking to model and optimize human trust and task performance in human-robot teams.

Sectors

Research and science · artificial intelligence and roboticsDefence and security · search-and-rescue mission

What the document contains

What's covered

The document covers a computational framework for balancing task performance and human trust in sequential human-robot interaction using potential-based reward shaping and an experience-based trust dynamics model.

What's missing

The text does not provide a necessary condition for guaranteeing small performance loss, nor does it evaluate potential functions other than linear forms.

Key takeaways

Maximizing task rewards in human-robot interaction often decreases human trust, whereas maximizing trust compromises performance.
The sequential human-robot interaction process can be effectively modeled as a two-player finite-horizon Markov game.
Potential-based reward shaping can guide a robot to increase human trust while mathematically bounding task performance loss.
Applying this framework to an experience-based trust model results in an efficiently solvable linear program.
Simulations of a search-and-rescue mission demonstrate that the proposed method successfully increases human trust at a minimal performance cost.

Outline

· I. Introduction
· II. Related Work
· III. Problem Formulation
· IV. Reward Shaping for Trustworthy Policy
· V. Case Study
· VI. Conclusion

Issued2 August 2023

Length6,385 words · 7 pages

FormatNot recorded

Scopeglobal

Physical AI Governancecontext

human_robot_interaction reinforcement_learning trust_modeling reward_shaping markov_games