Robot Policy Learning on DROID Franka Panda
47.4Average Success RateKT + Mix-Mask + FiLM
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| KT + Mix-Mask + FiLMKinematic Token=true, Mask=Mix-Mask, FiLM=true, Chunk size (G)=12026.02 | 47.4 | 5.7 | 77.7 | 58.7 | |
| KT + FiLMKinematic Token=true, Mask=None, FiLM=true, Chunk size (G)=12026.02 | 37.7 | 6 | 65 | 42 | |
| KT + Mix-MaskKinematic Token=true, Mask=Mix-Mask, FiLM=false, Chunk size (G)=12026.02 | 36.9 | 27.7 | 56.7 | 26.3 | |
| KTKinematic Token=true, Mask=None, FiLM=false, Chunk size (G)=12026.02 | 36 | 10.3 | 67.7 | 30 | |
| KT + Full-MaskKinematic Token=true, Mask=Full-Mask, FiLM=false, Chunk size (G)=12026.02 | 30.1 | 15.3 | 34 | 41 | |
| KT + Soft-MaskKinematic Token=true, Mask=Soft-Mask, FiLM=false, Chunk size (G)=12026.02 | 26.1 | 29.7 | 17.3 | 31.3 | |
| pi0.5 baselineKinematic Token=false, Mask=None, FiLM=false, Chunk size (G)=12026.02 | 19.7 | 18.3 | 15.7 | 25 |