Research map / Reinforcement learning / MDPs and planning
DSADF: Thinking Fast and Slow for Decision Making
Zhihao Dou, Dongfei Cui, Jun Yan, Weida Wang, Benteng Chen, Haoming Wang, Zeke Xie, Shufei Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add the large VLM planner (System 2) (ablation) | DSADF · Crafter in-domain Task 2 | TSR (%): 71.67 -> 88.33 | Improved |
| add the small VLM auxiliary performer (ablation) | DSADF · Crafter in-domain Task 2 | TSR (%): 81.67 -> 88.33 | Improved |
| add the RL agent (ablation) | DSADF · Crafter in-domain Task 2 | TSR (%): 38.33 -> 88.33 | Improved |
| add self-reflection (ablation) | DSADF · Crafter in-domain Task 2 | TSR (%): 80.00 -> 88.33 | Improved |
| add self-reflection (ablation) | DSADF · Crafter OOD Task 5 | TSR (%): 65.00 -> 68.33 | Improved |
| add self-reflection (ablation) | DSADF · Crafter OOD Task 5 | Time (s): 2884.7 -> 2767.5 | Improved |
| add the large VLM planner and auxiliary performer (ablation) | DSADF · Crafter OOD Task 5 | TSR (%): 0.00 -> 68.33 | Improved |
| use GPT-4o as planner instead of DeepSeek (ablation) | DSADF · Crafter OOD Task 5 | TSR (%): 65.50 -> 68.33 | Improved |
| use GPT-4o as planner instead of DeepSeek (ablation) | DSADF · Crafter OOD Task 6 | TSR (%): 77.50 -> 75.00 | Hurt |
| use DSADF instead of small VLM agents with GPT-4o planner | Qwen-2.5-VL/GPT-4o · Crafter OOD Task 5 | TSR (%): 21.67 -> 68.33 | Improved |
| use DSADF instead of GPT-4o as the agent | GPT-4o · Crafter OOD Task 8 | TSR (%): 55.00 -> 91.67 | Improved |
| use Qwen-2.5-VL as auxiliary performer instead of LLaVA-1.5 (ablation) | DSADF · Crafter OOD Task 4 | TSR (%): 63.33 -> 68.33 | Improved |
Closest papers by meaning
- BALROG: Benchmarking Agentic LLM and VLM Reasoning On Games (ICLR 2025 · similarity 0.88)
- Synthesizing Multimodal Verifiable Game Data to Boost VLMs' General Reasoning (ICLR 2026 · similarity 0.88)
- GFlowVLM: Enhancing Multi-step Reasoning in Vision-Language Models with Generative Flow Networks (CVPR 2025 · similarity 0.88)
- Fine-Tuning Large Vision-Language Models as Decision-Making Agents via Reinforcement Learning (NeurIPS 2024 · similarity 0.87)
- Think Twice, Act Once: A Co-Evolution Framework of LLM and RL for Large-Scale Decision Making (ICML 2025 · similarity 0.87)
- LLM-PySC2: Starcraft II learning environment for Large Language Models (NeurIPS 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
