# VARP: Reinforcement Learning from Vision-Language Model Feedback with Agent Regularized Preferences

arXiv: https://arxiv.org/abs/2503.13817 · HTML: https://mlsanity.com/paper/2503.13817/

## Key facts

- Cited by 9 papers in the map (data through 14 Oct 2026).
- Placed in Embodied and 3D vision-language models, one of 16 directions in Multimodal LLMs / vision-language models.
- Builds on 27 of its 32 references that are in the map.
- First posted on arXiv 18 Mar 2025.

## Placed in

- Multimodal LLMs / vision-language models / [Embodied and 3D vision-language models](https://mlsanity.com/mllm/embodied-and-3d-vision-language-models/)
- LLM agents and tool use / [Reinforcement learning for LLM agents](https://mlsanity.com/agents/reinforcement-learning-for-llm-agents/)
- LLM alignment / RLHF / preference optimisation / [Reward modeling and RLHF](https://mlsanity.com/alignment/reward-modeling-and-rlhf/)

## Builds on (most cited first, 27 of 27)

- [Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor](https://mlsanity.com/paper/1801.01290/) (2018)
- [Deep reinforcement learning from human preferences](https://mlsanity.com/paper/1706.03741/) (2017)
- [Concrete Problems in AI Safety](https://mlsanity.com/paper/1606.06565/) (2016)
- [Meta-World: A Benchmark and Evaluation for Multi-Task and Meta Reinforcement Learning](https://mlsanity.com/paper/1910.10897/) (2019)
- [DeepMind Control Suite](https://mlsanity.com/paper/1801.00690/) (2018)
- [Vision-Language Models for Vision Tasks: A Survey](https://mlsanity.com/paper/2304.00685/) (2023)
- [Habitat 2.0: Training Home Assistants to Rearrange their Habitat](https://mlsanity.com/paper/2106.14405/) (2021)
- [Scalable agent alignment via reward modeling: a research direction](https://mlsanity.com/paper/1811.07871/) (2018)
- [Reward learning from human preferences and demonstrations in Atari](https://mlsanity.com/paper/1811.06521/) (2018)
- [Language to Rewards for Robotic Skill Synthesis](https://mlsanity.com/paper/2306.08647/) (2023)
- [PEBBLE: Feedback-Efficient Interactive Reinforcement Learning via Relabeling Experience and Unsupervised Pre-training](https://mlsanity.com/paper/2106.05091/) (2021)
- [Reward Design with Language Models](https://mlsanity.com/paper/2303.00001/) (2023)
- [Defining and Characterizing Reward Hacking](https://mlsanity.com/paper/2209.13085/) (2022)
- [RL-VLM-F: Reinforcement Learning from Vision Language Foundation Model Feedback](https://mlsanity.com/paper/2402.03681/) (2024)
- [RT-Trajectory: Robotic Task Generalization via Hindsight Trajectory Sketches](https://mlsanity.com/paper/2311.01977/) (2023)
- [Reward (Mis)design for Autonomous Driving](https://mlsanity.com/paper/2104.13906/) (2021)
- [Preference-based Online Learning with Dueling Bandits: A Survey](https://mlsanity.com/paper/1807.11398/) (2018)
- [Contextual Dueling Bandits](https://mlsanity.com/paper/1502.06362/) (2015)
- [SURF: Semi-supervised Reward Learning with Data Augmentation for Feedback-efficient Preference-based Reinforcement Learning](https://mlsanity.com/paper/2203.10050/) (2022)
- [Multi-Objective Deep Reinforcement Learning](https://mlsanity.com/paper/1610.02707/) (2016)
- [PARL: A Unified Framework for Policy Alignment in Reinforcement Learning from Human Feedback](https://mlsanity.com/paper/2308.02585/) (2023)
- [RIME: Robust Preference-based Reinforcement Learning with Noisy Preferences](https://mlsanity.com/paper/2402.17257/) (2024)
- [Instructing Robots by Sketching: Learning from Demonstration via Probabilistic Diagrammatic Teaching](https://mlsanity.com/paper/2309.03835/) (2023)
- [TarGF: Learning Target Gradient Field to Rearrange Objects without Explicit Goal Specification](https://mlsanity.com/paper/2209.00853/) (2022)
- [Reinforcement Learning for Robotic Manipulation using Simulated Locomotion Demonstrations](https://mlsanity.com/paper/1910.07294/) (2019)
- [Sketch-to-Skill: Bootstrapping Robot Learning with Human Drawn Trajectory Sketches](https://mlsanity.com/paper/2503.11918/) (2025)
- [PLANRL: A Motion Planning and Imitation Learning Framework to Bootstrap Reinforcement Learning](https://mlsanity.com/paper/2408.04054/) (2024)

## Built on by (most cited first, 9 of 9)

- [Perception-Aware Policy Optimization for Multimodal Reasoning](https://mlsanity.com/paper/2507.06448/) (2025)
- [SOLE-R1: Video-Language Reasoning as the Sole Reward for On-Robot Reinforcement Learning](https://mlsanity.com/paper/2603.28730/) (2026)
- [Cross-Modal Instructions for Robot Motion Generation](https://mlsanity.com/paper/2509.21107/) (2025)
- [ExpAlign: Expectation-Guided Vision-Language Alignment for Open-Vocabulary Grounding](https://mlsanity.com/paper/2601.22666/) (2026)
- [VLM-Guided Experience Replay](https://mlsanity.com/paper/2602.01915/) (2026)
- [VersualRL: Closed-Loop Verbal Reinforcement Learning with Visual Execution Feedback for Task-Level Robot Planning](https://mlsanity.com/paper/2603.22169/) (2026)
- [HELP: Human-Efficient Large-Scale Robot Post-Training with Rollout Segmentation](https://mlsanity.com/paper/2607.09776/) (2026)
- [Beyond Pairwise Feedback: Listwise Vision-Language Supervision for Preference-Based Reward Learning](https://mlsanity.com/paper/2608.25350/) (2026)
- [Same Trajectory, Contradictory Rewards (ROBORMBENCH): Paraphrase Fragility in Vision Language Reward Models](https://mlsanity.com/paper/2609.05401/) (2026)

## Abstract (arXiv)

Designing reward functions for continuous-control robotics often leads to subtle misalignments or reward hacking, especially in complex tasks. Preference-based RL mitigates some of these pitfalls by learning rewards from comparative feedback rather than hand-crafted signals, yet scaling human annotations remains challenging. Recent work uses Vision-Language Models (VLMs) to automate preference labeling, but a single final-state image generally fails to capture the agent's full motion. In this paper, we present a two-part solution that both improves feedback accuracy and better aligns reward learning with the agent's policy. First, we overlay trajectory sketches on final observations to reveal the path taken, allowing VLMs to provide more reliable preferences-improving preference accuracy by approximately 15-20% in metaworld tasks. Second, we regularize reward learning by incorporating the agent's performance, ensuring that the reward model is optimized based on data generated by the current policy; this addition boosts episode returns by 20-30% in locomotion tasks. Empirical studies on metaworld demonstrate that our method achieves, for instance, around 70-80% success rate in all tasks, compared to below 50% for standard approaches. These results underscore the efficacy of combining richer visual representations with agent-aware reward regularization.