Mutual Adaptation in Human-Robot Co-Transportation with Human Preference Uncertainty

Al Jaber Mahmud, Weizi Li, Xuan Wang

Abstract

Mutual adaptation can significantly enhance overall task performance in human-robot co-transportation by integrating both the robot's and human's understanding of the environment. While human modeling helps capture humans' subjective preferences, two challenges persist: (i) the uncertainty of human preference parameters and (ii) the need to balance adaptation strategies that benefit both humans and robots. In this paper, we propose a unified framework to address these challenges and improve task performance through mutual adaptation. First, instead of relying on fixed parameters, we model a probability distribution of human choices by incorporating a range of uncertain human parameters. Next, we introduce a time-varying stubbornness measure and a coordination mode transition model, which allows either the robot to lead the team's trajectory or, if a human's preferred path conflicts with the robot's plan and their stubbornness exceeds a threshold, the robot to transition to following the human. Finally, we introduce a pose optimization strategy to mitigate the uncertain human behaviors when they are leading. To validate the framework, we design and perform experiments with real human feedback. We then demonstrate, through simulations, the effectiveness of our models in enhancing task performance with mutual adaptation and pose optimization.

Introduction

When a human and a robot carry a large object together, such as a table or a board, they must reach a consensus on the path to take. However, humans are not perfectly rational actors; we possess subjective biases, such as overestimating small risks or underestimating long distances. Furthermore, these preferences are not static—they vary between individuals and evolve as a task progresses.

A human and robot co-transporting an object in a laboratory setting.

Existing research often assumes fixed human parameters or relies on "single-direction" adaptation, where either the robot always follows or the human is forced to adapt to the robot's rigid plan. This paper addresses the challenge of mutual adaptation under uncertainty: how can a robot model a human's subjective, uncertain preferences and dynamically decide when to lead versus when to yield to the human’s choice?

Proposed Framework

The authors propose a three-part framework designed to bridge the gap between objective robotic planning and subjective human psychology.

Overview of the proposed mutual adaptation and pose optimization framework.

1. Probabilistic Human Preference Model

Instead of assuming a human has a single fixed cost function, the framework utilizes cognitive models—specifically Prelec’s function for risk perception and Steven’s power law for distance. By treating these parameters as probability distributions rather than constants, the robot can anticipate a range of likely human choices, making the system more robust to individual variability.

2. Time-Varying Stubbornness & Mode Transition

A novel "stubbornness" measure, $\phi(k)$, is introduced to track human discomfort during the task. If the robot leads the team toward a path the human dislikes, the human's stubbornness increases. When this value crosses a specific threshold, the robot transitions from a "leading" mode to a "following" mode, allowing for a dynamic shift in team hierarchy that respects human preference.

3. Pose Optimization in Low-Level Control

When the human takes the lead, the robot faces uncertainty regarding the exact path the human will choose. To mitigate this, the robot leverages its kinematic redundancy to optimize its pose. It positions its joints to minimize expected control effort across all likely human-chosen paths, effectively "hedging its bets" to avoid awkward configurations or singularities.

Key Insights and Results

  • Stubbornness as a Control Signal: Treating stubbornness as a mathematical state provides a principled way to handle the "who is in charge" problem in human-robot interaction (HRI).
  • Redundancy as an Uncertainty Buffer: Using pose optimization as a proactive tool allows the robot to remain flexible and ready for human-led decisions, significantly reducing control costs.
  • Performance Gains: Simulations demonstrated that this mutual adaptation model consistently resulted in lower objective costs for the team compared to passive adaptation baselines. For instance, in one environment, the true control cost dropped from approximately 1078 to 843 when pose optimization was active.

Conclusion and Future Work

This framework successfully demonstrates that modeling human uncertainty and allowing for dynamic leadership transitions improves co-transportation efficiency. Future work aims to expand this model to include multiple hand-offs of leadership and to account for additional human factors such as physical fatigue, stress, and trust levels. Additionally, the researchers plan to scale the system to multi-human, multi-robot scenarios, which will introduce more complex coordination challenges.

Summary updated

Abstract

Mutual adaptation can significantly enhance overall task performance in human-robot co-transportation by integrating both the robot's and human's understanding of the environment. While human modeling helps capture humans' subjective preferences, two challenges persist: (i) the uncertainty of human preference parameters and (ii) the need to balance adaptation strategies that benefit both humans and robots. In this paper, we propose a unified framework to address these challenges and improve task performance through mutual adaptation. First, instead of relying on fixed parameters, we model a probability distribution of human choices by incorporating a range of uncertain human parameters. Next, we introduce a time-varying stubbornness measure and a coordination mode transition model, which allows either the robot to lead the team's trajectory or, if a human's preferred path conflicts with the robot's plan and their stubbornness exceeds a threshold, the robot to transition to following the human. Finally, we introduce a pose optimization strategy to mitigate the uncertain human behaviors when they are leading. To validate the framework, we design and perform experiments with real human feedback. We then demonstrate, through simulations, the effectiveness of our models in enhancing task performance with mutual adaptation and pose optimization.
BibTeX
@misc{mahmud2025mutual,
  author = {Al Jaber Mahmud and Weizi Li and Xuan Wang},
  title = {{Mutual Adaptation in Human-Robot Co-Transportation with Human Preference Uncertainty}},
  year = {2025},
  eprint = {2503.08895},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2503.08895},
}
Figure