# D&A: Decoupling Description and Assessment Enables VLMs as Zero-Shot Robotic Reward Models

Source: [arXiv](https://arxiv.org/abs/2611.03622v1)  
Feed7 permalink: https://feed7.dev/p/2611-03622v1-0tys9eg  
Published: 2026-11-02T18:52:11.000Z  
Trust: Needs Review (needs_review)

## Why Included

D&A splits video progress scoring into scene description and rubric-based assessment, aiming to reduce VLMs’ tendency to mistake failed robot runs for progress.

## Source Summary

**D&A (Describe & Assess)** first produces a task-grounded description of scene dynamics, then scores progress against an explicit rubric. The authors report that this prompt-only method captures partial progress and penalizes failed robot executions.

## Practical Implication

For agents that evaluate video or tool outcomes, define what success means independently of the clip’s apparent trajectory, then assess evidence against that criterion.

## Agent-Ready Context

**D&A (Describe & Assess)** first produces a task-grounded description of scene dynamics, then scores progress against an explicit rubric. The authors report that this prompt-only method captures partial progress and penalizes failed robot executions.

For agents that evaluate video or tool outcomes, define what success means independently of the clip’s apparent trajectory, then assess evidence against that criterion.

The reported results concern zero-shot robotic task-progress estimates. The supplied material gives no numeric results or details about how performance varies by task, embodiment, or benchmark.

## Connected Context

Feed7 judgment across 990 accumulated Signals:

This gives a concrete video-evaluation pattern: describe task-relevant scene changes, then assess them against an explicit rubric so apparent progress does not substitute for success criteria. Dyna’s robotics account also emphasizes progress scoring and inspected trajectories, making scorer repeatability relevant; the supplied brief gives no numeric evidence about D&A’s performance.

- [Robot Demos Are Easy. Reliability Is Hard — Jason Ma, Dyna Robotics](https://feed7.dev/p/robot-demos-are-easy-reliability-is-hard-jason-ma-dyna-robotics-0h07ncw) — Dyna’s account uses reward models to detect rare failures and guide recovery, while D&A offers a rubric-based way to score progress and penalize failed executions; both make the quality of the evaluator central to robotics reliability.
- [How Evals and Prompts Shape Agent Behavior — Preetika Bhateja & Daniel Bump, YouTube Ads](https://feed7.dev/p/how-evals-and-prompts-shape-agent-behavior-preetika-bhateja-daniel-bump-1cmecaw) — The production-evaluation loop calls for calibrated judges and trace review; D&A specifies a way to ground a judge’s assessment in task criteria rather than a clip’s apparent trajectory.

## Context Map

- Layer: context
- Domains: research, video
- Topics: prompting, agent-evals

## Uncertainty

- The reported results concern zero-shot robotic task-progress estimates. The supplied material gives no numeric results or details about how performance varies by task, embodiment, or benchmark.

## Agent Instruction

Use this item as source-backed context. Do not invent claims beyond the linked source. If this item conflicts with another source, call out the conflict.