D&A: Decoupling Description and Assessment Enables VLMs as Zero-Shot Robotic Reward Models
cs.RO, cs.AI, cs.CL, cs.LG, cs.SY, eess.SY
Submitted: 2026-11-02
Updated: 2026-11-02
Project page: https://weirdlabuw.github.io/dna
Terminology
Sources
- Challenging the Evaluator: LLM Sycophancy Under User Rebuttal
- Good Arguments Against the People Pleasers: How Reasoning Mitigates (Yet Masks) LLM Sycophancy
- BrokenMath: A Benchmark for Sycophancy in Theorem Proving with LLMs
- TOPReward: Token Probabilities as Hidden Zero-Shot Rewards for Robotics
- A Vision-Language-Action-Critic Model for Robotic Real-World Reinforcement Learning
- Robo-Dopamine: General Process Reward Modeling for High-Precision Robotic Manipulation
- RoboReward: General-Purpose Vision-Language Reward Models for Robotics
- Robometer: Scaling General-Purpose Robotic Reward Models via Trajectory Comparisons
- RARM: Confidence-Gated Progress Reward Modeling for RL in Manipulation
- ReWiND: Language-Guided Rewards Teach Robot Policies without New Demonstrations
- $\pi^{*}_{0.6}$: a VLA That Learns From Experience
- RynnValue: Scaling Robotic Value Foundation Models with Temporal Distance
- PROGRESSLM: Towards Progress Reasoning in Vision-Language Models
- SOLE-R1: Video-Language Reasoning as the Sole Reward for On-Robot Reinforcement Learning
- OpenAI GPT-5 System Card
- FAST: Efficient Action Tokenization for Vision-Language-Action Models
- OpenGVL -- Benchmarking Visual Temporal Progress for Data Curation
- Rubrics as Rewards: Reinforcement Learning Beyond Verifiable Domains
- Steering Your Diffusion Policy with Latent Space Reinforcement Learning
- Diffusion Guidance Is a Controllable Policy Improvement Operator
Related papers
- Rethinking Actions: Continuity-Aware Priors for World-Model Planning
- MOSAIC: Modular Obstacle and Spatial-Aware Framework for Manipulation In Cluttered Scenes
- SAGAS: Semantic-Aware Graph-Assisted Stitching for Offline Temporal Logic Planning
- DexSeed: Expanding Human Hand-Object Interactions into Robot Demonstrations for Dexterous Policy Learning
- LaST0.5: A Generalist World Action Model with Interaction-Centric Latent Reasoning
- CIALLO: Learning Shared Intent Representations for Cross-Embodiment Manipulation