Robot Manipulation on CaP-Bench Robosuite (held-out)
70Overall Success RateRHO
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| RHOEvaluation setting=S4, Execution=single-turn, Visual-difference module=none2026.06 | 70 | 98 | 98 | 54 | 100 | 100 | 53 | 86 | |
| CaP-Agent0Evaluation setting=M4, Execution=multi-turn, Visual-difference module=enabled2026.06 | 68.29 | 97 | 98 | 89 | 0 | 100 | 74 | 20 | |
| CaP-Agent0Evaluation setting=S4, Execution=single-turn, Visual-difference module=none2026.06 | 24 | — | — | — | — | — | — | — |