Robot Manipulation on RLBench 10
100Success RateACT
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| ACTTask=open-drawer2026.05 | 100 | 90.8 | — | — | |
| SkiPTask=open-drawer2026.05 | 100 | 39.8 | — | — | |
| SkiPTask=stack-wine2026.05 | 100 | 77.3 | — | — | |
| SkiPTask=sweep-dustpan2026.05 | 100 | 59.4 | — | — | |
| SkiPTask=push-button2026.05 | 98 | 26.2 | — | — | |
| CoA-fwdTask=push-button2026.05 | 97 | 126.6 | — | — | |
| SkiPTask=take-lid-off2026.05 | 97 | 33.4 | — | — | |
| CoA-fwdTask=sweep-dustpan2026.05 | 96 | 120.2 | — | — | |
| SkiPTask=open-box2026.05 | 91 | 87.1 | — | — | |
| ACTTask=stack-wine2026.05 | 90 | 163.6 | — | — | |
| SkiPEvaluation Scope=Aggregate2026.05 | 85 | 72.9 | 43.4 | 1.6 | |
| KF-onlyTask=reach-target2026.05 | 72 | 37.2 | — | — | |
| ACTTask=reach-target2026.05 | 72 | 46.6 | — | — | |
| ACTEvaluation Scope=Aggregate2026.05 | 71 | 139.4 | 106.1 | 3.1 | |
| CoA-revEvaluation Scope=Aggregate2026.05 | 70 | 127.5 | 86.2 | 3.1 | |
| CoA-fwdEvaluation Scope=Aggregate2026.05 | 68 | 161.5 | 133.3 | 3.5 | |
| KF-onlyEvaluation Scope=Aggregate2026.05 | 49 | 113.1 | 10.9 | 4.3 | |
| DPEvaluation Scope=Aggregate2026.05 | 43 | 160 | 119.1 | 5.4 | |
| DPTask=open-box2026.05 | 34 | 201.5 | — | — | |
| SkiP†Evaluation Scope=Aggregate, variant=Gemini-2.5-Pro key segments2026.05 | 30 | 183.1 | 97.6 | 6.3 |