WatchAct - Error Correction: leaderboard
The Error Correction task of WatchAct (Li et al., 2026): after watching a real video of a person in a tabletop scene, the robot planner detects and corrects the person's mistakes. A vision-language model writes a high-level Pick/Place/Open/Close plan that is executed symbolically in an aligned LIBERO scene; plan success rate in percent over 240 instances, higher is better.
Metric: Plan success rate (%; the VLM's high-level Pick/Place/Open/Close plan, predicted from a real human video and an instruction, run through a symbolic executor and checked against the task's success predicate; 240 instances). Source: arxiv.org. Saturation forecast: Around April 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 43.2 | #91 |
| 2 | Gemini 3 Flash | 41.6 | #159 |
| 3 | GPT-5.4 | 37.6 | #121 |
| 4 | Qwen 3 VL 235B A22B (Thinking) | 36 | #451 |
| 5 | Gemini 3.1 Flash Lite | 34.4 | #363 |
| 6 | GPT-5.1 | 29.6 | #225 |
| 7 | Qwen 3 VL 235B A22B Instruct | 22.4 | #521 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=watchact-error-correction · How It Works · Data refreshed daily, snapshot 2026-11-12.