Embodied Planning on ALFWorld Seen v1 (in-distribution)
57.14Average Success RateProCeedSFT
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| ProCeedSFTBackbone=Qwen3-8B2026.04 | 57.14 | — | — | — | — | — | — | — | |
| ProCeedRLBackbone=Qwen3-8B2026.04 | 51.43 | — | — | — | — | — | — | — | |
| RFTBackbone=Qwen3-8B2026.04 | 47.38 | — | — | — | — | — | — | — | |
| DAPOBackbone=Qwen3-8B2026.04 | 45.23 | — | — | — | — | — | — | — | |
| Qwen3-8BBackbone=Qwen3-8B2026.04 | 44.22 | — | — | — | — | — | — | — | |
| GFlowVLM w/ SubTBAssump.=NM, SFT Initialization=true2025.03 | 26.1 | 50 | 23.1 | 10 | 18.7 | 24.3 | 23.7 | 1.4 | |
| GFlowVLM w/ Var-TBAssump.=NM, SFT Initialization=false2025.03 | 25.7 | 50 | 22.2 | 10.2 | 16.1 | 22.7 | 21.9 | 1.41 | |
| RL4VLMAssump.=NM, SFT Initialization=true2025.03 | 22.1 | 49.1 | 13.5 | 9.8 | 15.2 | 20.1 | 20.6 | 1.11 | |
| RL4VLMAssump.=M, SFT Initialization=true2025.03 | 21.7 | 47.4 | 14.7 | 10.4 | 14.4 | 18.8 | 18 | 1.12 | |
| ProCeedRLBackbone=Qwen3-1.7B2026.04 | 20.35 | — | — | — | — | — | — | — | |
| DAPOBackbone=Qwen3-1.7B2026.04 | 18.69 | — | — | — | — | — | — | — | |
| SFT-w/o-Assump.=-2025.03 | 17.1 | 39.2 | 0 | 14.4 | 11.1 | 0 | 28.6 | 1.06 | |
| SFT-w/-Assump.=-2025.03 | 15.9 | 32.7 | 0 | 10.3 | 10.8 | 0 | 21.8 | 1.02 | |
| Qwen3-1.7BBackbone=Qwen3-1.7B2026.04 | 11.07 | — | — | — | — | — | — | — |