| Task Name | Dataset Name | SOTA Result | Trend | |
|---|---|---|---|---|
| Multi-step manipulation | DROID Tabletop Multi-step tasks | Success Rate98 | 18 | |
| Semantic reasoning manipulation | DROID Tabletop Semantic tasks | Success Rate26 | 18 | |
| Rearrangement with distractors | DROID Tabletop Distractor tasks | Success Rate27 | 18 | |
| Video Generation | DROID Wrist | LPIPS0.364 | 16 | |
| Offline Action Matching | DROID | Action Score62.1 | 14 | |
| Predicate Detection | DROID phasic episodes (test) | F1 Score96.8 | 12 | |
| Pick-and-place | DROID Tabletop Simple tasks | Success Rate27 | 12 | |
| Classification | Droid (test) | Accuracy97.18 | 10 | |
| Physical Routing | DROID memory axis (physical evaluation) | Q (%)100 | 10 | |
| Video Generation | DROID (Exterior) | LPIPS0.117 | 8 | |
| Future state retrieval | DROID t+5 (2.0s prediction horizon) | Hit@142.97 | 8 | |
| Future state retrieval | DROID t+3 (1.2s prediction horizon) | Hit@148.27 | 8 | |
| Future state retrieval | DROID t+1 (0.4s prediction horizon) | Hit@154.11 | 8 | |
| Video Depth | DROID | Abs Rel0.223 | 8 | |
| Policy Identification | DROID Specialization | Probe Accuracy100 | 7 | |
| Robot Policy Learning | DROID Franka Panda | Average Success Rate47.4 | 7 | |
| Video-to-Video Generation | Droid (test) | VBench0.81 | 6 | |
| Interactive long-trajectory generation | DROID (val) | PSNR23.56 | 6 | |
| Video Frame Rank-Correlation | DROID | VOC Rank-Correlation (Sparse)0.99 | 6 | |
| Video Generation | DROID | PSNR24.37 | 5 | |
| Robotic Manipulation | DROID Sharpa Hand Unseen v1 (test) | Pick and Place Progression Rate93.8 | 5 | |
| Robotic Manipulation | DROID YAM Arm and Gripper Unseen v1 (test) | Pick and Place Progression Rate93.8 | 5 | |
| Robotic Manipulation | DROID UMI Gripper Unseen v1 (test) | Pick and Place Progression Rate97.9 | 5 | |
| Robotic Manipulation | DROID Original Gripper v1 (test) | Pick and Place Progression Rate100 | 5 | |
| Wipe Table with Towel | Real-world DROID platform | Success Rate95 | 5 |