Vision language models improve robot task progress measurement
D&A: Decoupling Description and Assessment Enables VLMs as Zero-Shot Robotic Reward Models
RoboticsArtificial IntelligenceComputation and LanguageMachine Learning
Summary
Measuring how well a robot is completing a task can be tricky, especially when the robot fails or makes partial progress. The authors introduce a new method called D&A, which makes the AI first describe what success looks like in words, then assess the robot's progress based on that description. This approach helps avoid mistakes where AI wrongly thinks the robot succeeded when it did not, and it works well without extra training. Their method can measure fine details of progress, helping robots learn better by themselves.
What this means in practice
- •For robotics engineers: Evaluate robot task progress precisely using scene descriptions without extra training data.
- •For automation developers: Design robots that learn from partial successes and failures more efficiently with text-based feedback.
Authors
Marius Memmel, Shawn Yin, Urvi Rutia, Angad Wadhwa, Eldon Li, Gavin Angeloff, Ankit Goyal, Jesse Zhang, Dieter Fox, Fabio Ramos, Anqi Li, Abhishek Gupta
Abstract
Vision-Language Models (VLMs) hold promise for zero-shot video task-progress estimation in robot learning, yet existing frameworks frequently suffer from confirmation bias, hallucinating progress in failed executions due to implicit success assumptions or relative temporal ranking designs. Naive prompting of frontier models collapses into a binary success detector that fails to capture nuanced, non-linear progress and failures. While post-training approaches can mitigate this through explicit training, they demand expensive real-world robotics data and require computationally intensive post-training for every individual base model. To resolve these limitations, we introduce D&A (Describe & Assess): our approach decouples progress estimation into an objective, task-grounded textual description of scene dynamics followed by an assessment phase using explicit rubrics. By compelling the VLM to synthesize an explicit, text-based definition of absolute success before scoring, the framework anchors evaluations to an external, global criterion rather than the video's internal timeline. Empirical results demonstrate that our method calibrates zero-shot estimates and mitigates confirmation bias, providing fine-grained, partial progress metrics that accurately penalize failed executions while correctly interpolating successful runs. Being purely prompt-based, D&A generalizes across tasks, embodiments, and benchmarks---matching or surpassing post-trained reward models---and its reward estimates enable efficient downstream reinforcement learning.