Low-Burden LLM-Based Preference Learning: Personalizing Assistive Robots from Natural Language Feedback for Users with Paralysis

2026-04-01

AI Summary

This work addresses the high physical and cognitive burden imposed on users with severe motor impairments by traditional preference learning methods, such as pairwise comparisons. To mitigate this, the authors propose a low-burden offline framework that directly translates unstructured natural language feedback from users into interpretable, deterministic robot control policies. The approach uniquely integrates the Occupational Therapy Practice Framework (OTPF) to guide a large language model (LLM) in parsing subjective user feedback and generating decision-tree-based policies. Additionally, an LLM-as-a-Judge mechanism is introduced to automatically verify policy safety. Evaluated in a simulated meal-preparation task, the system significantly reduces user workload, and independent clinical experts confirmed that the resulting policies are safe, transparent, and accurately reflect user preferences.

Abstract

Physically Assistive Robots (PARs) require personalized behaviors to ensure user safety and comfort. However, traditional preference learning methods, like exhaustive pairwise comparisons, cause severe physical and cognitive fatigue for users with profound motor impairments. To solve this, we propose a low-burden, offline framework that translates unstructured natural language feedback directly into deterministic robotic control policies. To safely bridge the gap between ambiguous human speech and robotic code, our pipeline uses Large Language Models (LLMs) grounded in the Occupational Therapy Practice Framework (OTPF). This clinical reasoning decodes subjective user reactions into explicit physical and psychological needs, which are then mapped into transparent decision trees. Before deployment, an automated "LLM-as-a-Judge" verifies the code's structural safety. We validated this system in a simulated meal preparation study with 10 adults with paralysis. Results show our natural language approach significantly reduces user workload compared to traditional baselines. Additionally, independent clinical experts confirmed the generated policies are safe and accurately reflect user preferences.

Problem

• preference learning
• assistive robots
• natural language feedback
• motor impairments
• user burden

Innovation

• Low-Burden Preference Learning
• Natural Language Feedback
• Large Language Models (LLMs)
• Occupational Therapy Practice Framework (OTPF)
• Interpretable Decision Trees