Robotic Planning on LEMMA Single-Agent Overall
80Success Rate (SR)SG-CoT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SG-CoTLLM=Gemini-2.5-Flash2026.03 | 80 | 78 | |
| InnerMonoLLM=Gemini-2.5-Flash2026.03 | 76 | 68 | |
| SG-CoTLLM=Qwen3-VL-2B2026.03 | 72 | 66 | |
| InnerMonoLLM=Qwen3-VL-2B2026.03 | 65 | 55 | |
| SG-CoTLLM=Gemini-2.5-Flash, Iterations=12026.03 | 63 | 53 | |
| SG-CoTLLM=Gemini-2.5-Flash, Scene Graph Edges=false2026.03 | 63 | 44 | |
| ProgPromptLLM=Gemini-2.5-Flash2026.03 | 55 | 45 | |
| SG-CoTLLM=Qwen3-VL-2B, Scene Graph Edges=false2026.03 | 55 | 43 | |
| SG-CoTLLM=Qwen3-VL-2B, Iterations=12026.03 | 54 | 46 | |
| CLARALLM=Gemini-2.5-Flash2026.03 | 52 | 49 | |
| SG-CoTLLM=Gemini-2.5-Flash, Iterations=02026.03 | 48 | 39 | |
| CLARALLM=Qwen3-VL-2B2026.03 | 44 | 40 | |
| SG-CoTLLM=Qwen3-VL-2B, Iterations=02026.03 | 38 | 28 | |
| ProgPromptLLM=Qwen3-VL-2B2026.03 | 37 | 32 |