Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Describe, Explain, Plan and Select: Interactive Planning with LLMs Enables Open-World Multi-Task Agents
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 describe/explain/re-plan loop (DEP) over best baseline planner | LLM planner · Minecraft Task101 avg | success rate (%): 25.77 -> 39.36 | Helped |
| add horizon-predictive selector to DEP (ablation) | DEPS · Minecraft Task101 avg | success rate (%): 39.36 -> 48.56 | Helped |
| increase re-planning rounds from 0 to unlimited (ablation) | DEPS · Minecraft MT1 | success rate (%): 28.6 -> 79.8 | Helped |
| increase re-planning rounds from 0 to unlimited (ablation) | DEPS · Minecraft MT7 | success rate (%): 0.6 -> 12.6 | Helped |
| increase re-planning rounds from 0 to unlimited (ablation) | DEPS · Minecraft MT8 ObtainDiamond | success rate (%): 0.0 -> 0.6 | Helped |
| use horizon-predictive selector instead of fixed goal order (ablation) | DEP · Minecraft parallel goals (2,3,4) | success rate: +22.3%, +29.2%, +32.6% | Helped |
| use horizon-predictive selector instead of CLIP/MineCLIP (ablation) | DEP · Minecraft parallel goals | success rate | Helped |
| add re-planning to zero-shot planner (GPT+RP vs GPT) | GPT planner · ALFWorld average | success rate (%): 10.0% -> 52.0% | Helped |
| add description and self-explanation (DEP vs GPT+RP) (ablation) | GPT+RP · ALFWorld average | success rate (%): 52.0% -> 76.0% | Helped |
| use DEP instead of GPT zero-shot planner | Tabletop Manipulation · Tabletop average | success rate (%): 80.0% -> 80.0% | No clear effect |
| run DEPS on MineRL with different controller | DEPS · Minecraft MT1-MT8 | success rate (%): MT3 62.40 -> 24.25; MT8 0.59 -> 1.84 | Mixed |
Closest papers by meaning
- MineAnyBuild: Benchmarking Spatial Planning for Open-world AI Agents (NeurIPS 2025 · similarity 0.90)
- MP5: A Multi-modal Open-ended Embodied System in Minecraft via Active Perception (CVPR 2024 · similarity 0.89)
- Experience-based Knowledge Correction for Robust Planning in Minecraft (ICLR 2026 · similarity 0.89)
- Optimus-1: Hybrid Multimodal Memory Empowered Agents Excel in Long-Horizon Tasks (NeurIPS 2024 · similarity 0.88)
- OmniJARVIS: Unified Vision-Language-Action Tokenization Enables Open-World Instruction Following Agents (NeurIPS 2024 · similarity 0.88)
- Craftium: Bridging Flexibility and Efficiency for Rich 3D Single- and Multi-Agent Environments (ICML 2025 · similarity 0.88)
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.