Marshmallows on Real-Robot
65Success RateBPP
Evaluation results
| Method | Links | |
|---|---|---|
| BPPPolicy Architecture=Diffusion Transformer, Training Objective=DDPM, Action Chunking=50, History Conditioning=Big Picture Policies2026.02 | 65 | |
| Current ObsPolicy Architecture=Diffusion Transformer, Training Objective=DDPM, Action Chunking=50, History Conditioning=Current Observation2026.02 | 40 | |
| PTPPolicy Architecture=Diffusion Transformer, Training Objective=DDPM, Action Chunking=50, History Conditioning=Past-Token Prediction2026.02 | 35 | |
| Naïve HistoryPolicy Architecture=Diffusion Transformer, Training Objective=DDPM, Action Chunking=50, History Conditioning=Naïve History2026.02 | 25 |