OGBench
Benchmarks
| Task Name | Dataset Name | SOTA Result | Trend | |
|---|---|---|---|---|
| Offline Reinforcement Learning | OGBench antmaze-large-navigate-singletask task1-v0 to task5-v0 | Score95.6 | 55 | |
| Offline Reinforcement Learning | OGBench | AntMaze Large Navigate95.9 | 27 | |
| Offline Reinforcement Learning | OGBench antmaze-giant-navigate-singletask task1-v0 to task5-v0 | Score52.2 | 22 | |
| Robot goal-reaching success rate evaluation | OGBench puzzle-4x4-play-sparse-singletask | Success Rate30 | 13 | |
| Robot goal-reaching success rate evaluation | OGBench puzzle-3x3-play-sparse-singletask | Success Rate100 | 13 | |
| Robot goal-reaching success rate evaluation | OGBench scene-play-sparse-singletask | Success Rate87 | 13 | |
| Robot goal-reaching success rate evaluation | OGBench cube-double-play-singletask | Success Rate (%)41 | 13 | |
| Robot goal-reaching success rate evaluation | OGBench cube-single-play-singletask | Success Rate99 | 13 | |
| Goal-conditioned Manipulation | OGBench puzzle-4x4-play | Score58 | 12 | |
| Goal-conditioned Manipulation | OGBench cube-double-play | Score79 | 12 | |
| Goal-conditioned Navigation | OGBench antsoccer-arena-navigate | Score77 | 12 | |
| Goal-conditioned Navigation | OGBench antmaze-giant-navigate | Score96 | 12 | |
| Goal-conditioned Navigation | OGBench antmaze-large-navigate | Score99 | 12 | |
| Goal-conditioned Navigation | OGBench humanoidmaze-large-navigate | Score33 | 12 | |
| Goal-conditioned Navigation | OGBench humanoidmaze-medium-navigate | Score97 | 12 | |
| scene-play | OGBench 100% offline dataset | Success Rate81 | 12 | |
| cube-single-play | OGBench 100% offline dataset | Success Rate0.7 | 12 | |
| antsoccer-arena-navigate | OGBench 100% offline | Success Rate73 | 12 | |
| antsoccer-medium-navigate | OGBench 100% offline dataset | Success Rate14 | 12 | |
| antmaze-medium-navigate | OGBench 100% offline dataset | Success Rate96 | 12 | |
| Offline Reinforcement Learning | OGBench Visual | Visual Cube Single Task 1 Success Rate92.7 | 11 | |
| Offline Reinforcement Learning | OGBench Average (50 tasks) | Score51.8 | 11 | |
| Offline Reinforcement Learning | OGBench cube-triple-play | Success Rate14 | 10 | |
| Singletask Offline Reinforcement Learning (State-based) | OGBench State-based Singletask Offline v0 | Success Rate97 | 10 | |
| Goal-conditioned Reinforcement Learning | OGBench scene play (5 tasks) zero-shot | Average Return19 | 10 |
Showing 25 of 72 rows