Robotic Manipulation on Unseen Position
30Unscrew Cap Success RateLaST0 (w/ unseen HD)
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| LaST0 (w/ unseen HD)Training Setting=Trained with additional human data in unseen scenarios, Zero-shot Evaluation=false2026.06 | 30 | 40 | 50 | 35 | 20 | 20 | 33 | 46 | |
| LaST-HD (w/ unseen HD)Training Setting=Trained with additional human data in unseen scenarios, Zero-shot Evaluation=false2026.06 | 30 | 40 | 60 | 45 | 35 | 35 | 41 | 56 | |
| LaST-HD (Mix-HD)Training Setting=Trained on in-domain data only, Zero-shot Evaluation=true2026.06 | 20 | 15 | 25 | 10 | 10 | 10 | 15 | 31 | |
| LaST0Training Setting=Trained on in-domain data only, Zero-shot Evaluation=true2026.06 | 15 | 10 | 35 | 5 | 20 | 5 | 15 | 30 | |
| π0.5Training Setting=Trained on in-domain data only, Zero-shot Evaluation=true2026.06 | 10 | 10 | 15 | 15 | 10 | 10 | 12 | 30 | |
| Cosmos-PolicyTraining Setting=Trained on in-domain data only, Zero-shot Evaluation=true2026.06 | 10 | 15 | 35 | 0 | 10 | 5 | 13 | 26 |