Reducing Oracle Feedback with Vision�Language Embeddings for Preference-Based RL
Udita Ghosh, Dripta Raychaudhuri, Jiachen Li, Konstantinos Karydis, Amit Roy-Chowdhury
AI summary
Problem
Preference-based reinforcement learning requires costly and time-consuming oracle feedback to learn reward functions, which limits its scalability and practical deployment in robotics.
Approach
ROVED combines lightweight vision-language embeddings for scalable preference generation with uncertainty-aware filtering to query oracles only when necessary, while continuously fine-tuning the model with sparse feedback.
Key results
- Matches oracle-only performance on eight Meta-World manipulation tasks
- Reduces oracle queries by 50–80% per task
- Achieves 75–90% cumulative annotation savings via cross-task VLE generalization
- Introduces parameter-efficient fine-tuning with dynamics-aware objectives
Why it matters
Enables scalable, cost-effective preference-based reward learning for real-world robotic applications where human or expensive model feedback is a bottleneck.
Abstract
Preference-based reinforcement learning can learn effective reward functions from comparisons, but its scalability is constrained by the high cost of oracle feedback. Lightweight vision-language embedding (VLE) models provide a cheaper alternative, but their noisy outputs limit their effectiveness as standalone reward generators. To address this challenge, we propose ROVED, a hybrid framework that combines VLE-based supervision with targeted oracle feedback. Our method uses the VLE to generate segment-level preferences and defers to an oracle only for samples with high uncertainty, identified through a filtering mechanism. In addition, we introduce a parameter-efficient fine-tuning method that adapts the VLE with the obtained oracle feedback in order to improve the model over time in a synergistic fashion. This ensures the retention of the scalability of embeddings and the accuracy of oracles, while avoiding their inefficiencies. Across multiple robotic manipulation tasks, ROVED matches or surpasses prior preference-based methods while reducing oracle queries by up to 80%. Remarkably, the adapted VLE generalizes across tasks, yielding cumulative annotation savings of up to 90%, highlighting the practicality of combining scalable embeddings with precise oracle supervision for preference-based RL. Project page: https://roved-icra-2026.github.io/