Bimanual Robot Manipulation on TWIN 1.0 (test)
99Push Box Success RateBiCICLe
Evaluation results
| Method | Links | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| BiCICLemode=training-free (ICL)2026.04 | 99 | 100 | 80.3 | 94.3 | 68 | 83.7 | 83 | 29 | 65.3 | 34.3 | 97.3 | 36 | 46.7 | 70.5 | |
| BiCICLe + Best-of-Nmode=training-free (ICL)2026.04 | 97.7 | 100 | 81.7 | 94 | 73.7 | 85 | 84.7 | 29.3 | 72.7 | 31.7 | 98.3 | 33.7 | 42.3 | 71.1 | |
| BiCICLe + Arms’ Debatemode=training-free (ICL)2026.04 | 97 | 100 | 83.3 | 94.3 | 70.3 | 80 | 83 | 28 | 65.3 | 33.3 | 96 | 36 | 47.3 | 70.3 | |
| PPImode=supervised2026.04 | 96.7 | — | — | — | 62.7 | 89.3 | 92 | 46.3 | — | — | 98.7 | — | 79.7 | — | |
| RoboPrompt-SAmode=training-free (ICL), architecture=Single-Agent (SA)2026.04 | 94 | 100 | 82 | 83.7 | 64.7 | 78.7 | 58.7 | 18 | 44 | 11.7 | 89.3 | 26.3 | 37.3 | 60.6 | |
| π0-keyposemode=supervised2026.04 | 93 | 38 | 22 | 2 | 59 | 97 | 72 | 27 | 41 | 7 | 2 | 68 | 40 | 43.7 | |
| 3DFAmode=supervised2026.04 | 92.7 | 92.7 | 89.3 | 89 | 96 | 99.7 | 94.7 | 74 | 69.7 | 40.7 | 99.3 | 94.7 | 93 | 85.1 | |
| KStarDiffusermode=supervised2026.04 | 83 | — | — | — | 27 | 98.7 | — | 43.7 | — | — | 89 | — | — | — | |
| RoboPrompt-DAmode=training-free (ICL), architecture=Multi-Agent (DA)2026.04 | 83 | 100 | 80 | 82.7 | 63.3 | 69.3 | 79.3 | 24.3 | 61 | 23.3 | 93.7 | 30 | 47 | 64.4 | |
| KAT-SAmode=training-free (ICL), architecture=Single-Agent (SA)2026.04 | 60 | 4.3 | 11.7 | 4 | 33.7 | 42.3 | 17 | 2.3 | 3.7 | 2.7 | 55.3 | 10 | 16.3 | 20.3 | |
| VLM-LFmode=training-free (ICL)2026.04 | 59 | 6.3 | 7 | 6.7 | 8.3 | 11.3 | 8 | 5.7 | 7.7 | 3 | 31 | 10 | 10 | 13.4 | |
| PerAct-LFmode=supervised2026.04 | 57 | 10 | 0 | 0 | 9 | 40 | 14 | 11 | 2 | 21 | 28 | 8 | 27 | 17.5 | |
| DP3mode=supervised2026.04 | 56 | — | — | — | 0 | 64 | — | 6.3 | — | — | 1.7 | — | — | — | |
| RVT-LFmode=supervised2026.04 | 52 | 39 | 0 | 0 | 0 | 17 | 6 | 3 | 3 | 3 | 0 | 3 | 10 | 10.5 | |
| KAT-DAmode=training-free (ICL), architecture=Multi-Agent (DA)2026.04 | 50.7 | 3 | 17.3 | 4 | 30.7 | 19 | 11.3 | 0.3 | 2.3 | 1.7 | 74.3 | 6 | 6 | 17.4 | |
| AnyBimanualmode=supervised2026.04 | 46 | 73 | 26 | 15 | 44 | 36 | 14 | 7 | 8 | 24 | 67 | 24 | — | 32 | |
| PerAct2mode=supervised2026.04 | 6 | 47 | 3 | 11 | 41 | 50 | 1 | 12 | 4 | 24 | 0 | 9 | 10 | 16.8 | |
| ACTmode=supervised2026.04 | 0 | 4 | 0 | 0 | 0 | 36 | 6 | 0 | 0 | 16 | 0 | 2 | 13 | 5.9 |