Robotic Policy Learning on Unitree G1 Dex1 (test)
28Average Success RateEmbodiment-aware Transformer (Full)
Evaluation results
| Method | Links | |
|---|---|---|
| Embodiment-aware Transformer (Full)Kinematic Token=✓, Mask=Mix-Mask, FiLM=✓, Chunk Size (G)=12026.02 | 28 | |
| Mix-Mask VariantKinematic Token=✓, Mask=Mix-Mask, FiLM=None, Chunk Size (G)=12026.02 | 27.3 | |
| FiLM VariantKinematic Token=✓, Mask=None, FiLM=✓, Chunk Size (G)=12026.02 | 25.3 | |
| Baseline TransformerKinematic Token=None, Mask=None, FiLM=None, Chunk Size (G)=12026.02 | 24.7 | |
| Kinematic Token VariantKinematic Token=✓, Mask=None, FiLM=None, Chunk Size (G)=12026.02 | 24.3 | |
| Full-Mask VariantKinematic Token=✓, Mask=Full-Mask, FiLM=None, Chunk Size (G)=12026.02 | 23.8 |