Robot manipulation via imitation learning on LIBERO Simulation Benchmark original (test)
84.7LIBERO Spatial SROpenVLA
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| OpenVLATraining=Fine-tuned, Adaptation=LoRA (r=32), Camera=Static third-person2024.06 | 84.7 | 1 | 88.4 | 2 | 79.2 | 2 | 53.7 | 1 | 76.5 | 1.5 | |
| OctoTraining=Fine-tuned, Camera=Static third-person2024.06 | 78.9 | 2 | 85.7 | 3 | 84.6 | 1 | 51.1 | 2 | 75.1 | 2 | |
| Diffusion PolicyTraining=From scratch, Camera=Static third-person2024.06 | 78.3 | 3 | 92.5 | 1 | 68.3 | 3 | 50.5 | 3 | 72.4 | 2.5 |