Robotic Task Planning and Code Generation on Basic
100Success Rate (SR)GSCE
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GSCE2026.04 | 100 | 100 | |
| Corrective GSCE2026.04 | 100 | 100 | |
| GSCE2026.04 | 100 | 100 | |
| Corrective GSCE2026.04 | 100 | 100 | |
| Ro-SLMGRPO optimization=none2026.04 | 97.7 | 98.9 | |
| Ro-SLM2026.04 | 97.7 | 98.9 | |
| Ro-SLMcomments=retained2026.04 | 95.5 | 96.6 | |
| Ro-SLMaugmentation=none2026.04 | 95.5 | 96.6 | |
| Ro-SLMvariant=without GRPO Optimization2026.04 | 93.2 | 95.5 | |
| Ro-SLM2026.04 | 93.2 | 95.5 | |
| Ro-SLMtraining tasks=Simple Task Only2026.04 | 90.9 | 93.2 | |
| Ro-SLMvariant=with Comments2026.04 | 90.9 | 93.2 | |
| Ro-SLMvariant=Simple Task Only2026.04 | 90.9 | 94.3 | |
| Ro-SLMvariant=without Augmentation2026.04 | 88.6 | 92 | |
| Llama-3.1-8B2026.04 | 11.4 | 26.1 | |
| Llama-3.1-8B2026.04 | 9.1 | 9.1 | |
| Ro-SLMtraining tasks=Complex Task Only2026.04 | 0 | 2.3 | |
| Ro-SLMvariant=Complex Task Only2026.04 | 0 | 2.3 |