Embodied household manipulation on ALFWorld Unseen domain
94Success RateSkillMAS
Evaluation Results
| Method | Links | |
|---|---|---|
| SkillMASEvaluation Source=Internal evaluation2026.05 | 94 | |
| Traj-BootstrapEvaluation Source=Published reference (contextual)2026.05 | 93 | |
| CDMemEvaluation Source=Published reference (contextual)2026.05 | 85.8 | |
| GRPOFramework=Agent-R1, Backbone=Qwen3-4B, Type=RL-optimized2025.11 | 74.58 | |
| RLOOFramework=Agent-R1, Backbone=Qwen3-4B, Type=RL-optimized2025.11 | 73.46 | |
| PPOFramework=Agent-R1, Backbone=Qwen3-4B, Type=RL-optimized2025.11 | 72.38 | |
| Reinforce++Framework=Agent-R1, Backbone=Qwen3-4B, Type=RL-optimized2025.11 | 69.57 | |
| ReActEvaluation Source=Published reference (contextual)2026.05 | 53 | |
| Direct LLMEvaluation Source=Published reference (contextual)2026.05 | 43.3 | |
| ReActFramework=Agent-R1, Backbone=Qwen3-4B, Type=Training-free baseline2025.11 | 2.98 |