REVIEW 9 cited by
Vision-Language Models are Zero-Shot Reward Models for Reinforcement Learning
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
Reinforcement learning (RL) requires either manually specifying a reward function, which is often infeasible, or learning a reward model from a large amount of human feedback, which is often very expensive. We study a more sample-efficient alternative: using pretrained vision-language models (VLMs) as zero-shot reward models (RMs) to specify tasks via natural language. We propose a natural and general approach to using VLMs as reward models, which we call VLM-RMs. We use VLM-RMs based on CLIP to train a MuJoCo humanoid to learn complex tasks without a manually specified reward function, such as kneeling, doing the splits, and sitting in a lotus position. For each of these tasks, we only provide a single sentence text prompt describing the desired task with minimal prompt engineering. We provide videos of the trained agents at: https://sites.google.com/view/vlm-rm. We can improve performance by providing a second "baseline" prompt and projecting out parts of the CLIP embedding space irrelevant to distinguish between goal and baseline. Further, we find a strong scaling effect for VLM-RMs: larger VLMs trained with more compute and data are better reward models. The failure modes of VLM-RMs we encountered are all related to known capability limitations of current VLMs, such as limited spatial reasoning ability or visually unrealistic environments that are far off-distribution for the VLM. We find that VLM-RMs are remarkably robust as long as the VLM is large enough. This suggests that future VLMs will become more and more useful reward models for a wide range of RL applications.
Forward citations
Cited by 9 Pith papers
-
SOLE-R1: Video-Language Reasoning as the Sole Reward for On-Robot Reinforcement Learning
A video-language model with per-timestep spatiotemporal CoT and dense progress prediction can serve as the sole reward for zero-shot online robot RL on 24 unseen manipulation tasks.
-
Weights or Skills? A Survey of Robot-Learning Techniques: from Action-Predicting Weights to Robots that Write their Own Skills
A taxonomy of robot learning on a weights-versus-skills axis, with a five-rung self-improvement ladder whose top cell (feedback plus memory plus search) holds only a few recent systems.
-
MARS-RA: Rank Aggregation for Credit Assignment via Multimodal Comparisons in Embodied Multi-Agent Cooperation
Credit assignment via LMM pairwise comparisons plus Bradley–Terry rank aggregation and potential-based shaping improves cooperative MARL under sparse rewards and dynamic agent counts.
-
PhyAgentOS: A Self-Evolving Operating System for Embodied Agents with Decoupled Cognitive Planning and Physical Execution
A file-based operating-system layer with a session verifier and persistent memory improves embodied-agent task completion on game, simulated, and real-robot platforms without retraining policies.
-
TOPReward: Token Probabilities as Hidden Zero-Shot Rewards for Robotics
The log-probability a VLM assigns to 'True' for 'does this video prefix complete the task?' is used as a zero-shot dense progress reward that outperforms GVL on open-source models.
-
FOUNDER: Grounding Foundation Models in World Models for Open-Ended Embodied Decision Making
FOUNDER maps foundation-model embeddings of text or video prompts into world-model goal states and rewards policies by predicted temporal distance to those goals, improving reward-free multi-task offline control.
-
RobotSmith: Generative Robotic Tool Design for Acquisition of Complex Manipulation Skills
RobotSmith autonomously designs, 3D-prints, and uses task-specific tools for robotic manipulation, raising task success from 2.8% (no tool) to 50% in simulation.
-
LLM-as-a-Verifier: A General-Purpose Verification Framework
Expecting over scoring-token logits yields continuous, scalable verification that improves agent trajectory selection and dense RL rewards across coding, robotics, and medical benchmarks.
-
Advancing Autonomous VLM Agents via Variational Subgoal-Conditioned Reinforcement Learning
VSC-RL combines VLM-generated subgoals with a subgoal-conditioned AWR-style RL objective and claims improved sample efficiency over DigiRL and WebRL on AitW and WebArena-Lite.
Discussion (0). Continue with ORCID to comment.