Embodied AI Planning on TDW-MAT
90.23Success RateBaseline
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| BaselineModel Scale=GPT-5, Execution Strategy=Baseline2026.04 | 90.23 | 41.34 | 5.8 | 40.5 | |
| ParallelModel Scale=GPT-5, Execution Strategy=Parallel2026.04 | 89.32 | 43.67 | 10.4 | 71.9 | |
| AgenticCache (Ours+)Model Scale=GPT-5, Execution Strategy=AgenticCache (Ours+)2026.04 | 88.64 | 22.27 | 4.1 | 27.7 | |
| BaselineModel Scale=GPT-5-mini, Execution Strategy=Baseline2026.04 | 85.45 | 29.6 | 4.5 | 5.4 | |
| ParallelModel Scale=GPT-5-mini, Execution Strategy=Parallel2026.04 | 84.55 | 30.84 | 7.5 | 8.9 | |
| AgenticCache (Ours+)Model Scale=GPT-5-mini, Execution Strategy=AgenticCache (Ours+)2026.04 | 84.32 | 22.56 | 3.6 | 4.2 | |
| SpeculativeModel Scale=GPT-5, Execution Strategy=Speculative2026.04 | 80.91 | 36.37 | 13.5 | 39.9 | |
| SpeculativeModel Scale=GPT-5-mini, Execution Strategy=Speculative2026.04 | 77.73 | 37.61 | 12.7 | 7.7 | |
| BaselineModel Scale=GPT-5-nano, Execution Strategy=Baseline2026.04 | 71.59 | 40.2 | 8.8 | 2.9 | |
| ParallelModel Scale=GPT-5-nano, Execution Strategy=Parallel2026.04 | 71.59 | 38.07 | 15.3 | 5 | |
| AgenticCache (Ours+)Model Scale=GPT-5-nano, Execution Strategy=AgenticCache (Ours+)2026.04 | 67.95 | 24.62 | 7.6 | 2.5 |