Robot Policy Evaluation on LIBERO-Plus Spatial
100Success Rate (Bg.)X-VLA
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| X-VLA2026.06 | 100 | 24.5 | 82.1 | — | 93.8 | 91.4 | 63.5 | 77.7 | |
| π02026.06 | 95 | 70.7 | 67.9 | — | 94 | 49.1 | 87.7 | 78.6 | |
| A2World-policyPre-training Strategy=T-pre (text-conditioned world-model pretraining)2026.06 | 89.9 | 73.1 | 93.3 | — | 94 | 92.9 | 60.7 | 85.8 | |
| Cosmos Policy2026.06 | 89.1 | 75.5 | 94.4 | — | 94.3 | 53.1 | 96 | 85.6 | |
| GE-Act2026.06 | 89.1 | 72.3 | 90.8 | — | 89.6 | 75.7 | 97.7 | 87.5 | |
| A2World-policyPre-training Strategy=P-pre (policy-targeted text-video-action pretraining)2026.06 | 89.1 | 75.8 | 100 | — | 92.2 | 91.4 | 72.9 | 88.6 | |
| A2World-policyPre-training Strategy=A-pre (action-to-video A2World pretraining)2026.06 | 88 | 74.2 | 100 | — | 94.5 | 93.1 | 80.9 | 88.5 | |
| A2World-policyPre-training Strategy=C-init (directly initializing from Cosmos-Predict2)2026.06 | 83.3 | 53.2 | 90.5 | — | 89.9 | 77.1 | 74.1 | 80.2 | |
| FastWAM2026.06 | 65.9 | 7.4 | 72.1 | — | 59.5 | 42.3 | 32.8 | 51.5 |