Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Grounded Decoding: Guiding Text Generation with Grounded Models for Embodied Agents
Wenlong Huang, Fei Xia, Dhruv Shah, Danny Driess, Andy Zeng, Yao Lu, Pete Florence, Igor Mordatch, Sergey Levine, Karol Hausman, brian ichter
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 grounded decoding (greedy) to LLM planner | LLM planner + CLIPort · Tabletop unseen Letters | success rate: 19% -> 37% | Improved |
| add grounded decoding (greedy) to LLM planner | LLM planner + CLIPort · Tabletop seen Box Packing | success rate: 11% -> 78% | Improved |
| use beam search instead of greedy decoding (ablation) | GD · Tabletop seen Blocks & Bowls | success rate: 60% -> 77% | Improved |
| use beam search instead of greedy decoding (ablation) | GD · Tabletop seen Box Packing | success rate: 79 -> 78% | No clear effect |
| use beam search instead of greedy decoding (ablation) | GD · Tabletop unseen Blocks & Bowls | success rate: 44% -> 50% | Improved |
| plan with LLM instead of end-to-end CLIPort (Long) | CLIPort · Tabletop unseen Letters | success rate: 10% -> 19% | Improved |
| add grounded decoding (greedy) to LLM planner | LLM planner + PPO skills · 2D Maze Hard | success rate: 54% -> 78% | Improved |
| use beam search instead of greedy decoding (ablation) | GD · 2D Maze Hard | success rate: 78% -> 88% | Improved |
| use beam search instead of greedy decoding (ablation) | GD · 2D Maze Easy | success rate: 100% -> 100% | No clear effect |
| hierarchical LLM planning instead of hierarchical RL | HRL · 2D Maze Medium | success rate: 48% -> 87% | Improved |
| use GD with open-vocabulary detector chain-of-thought instead of SayCan | SayCan · Kitchen ambiguous | planning success rate: 33% -> 58% | Improved |
| use GD instead of SayCan on unambiguous tasks | SayCan · Kitchen unambiguous | planning success rate: 85% -> 85% | No clear effect |
Closest papers by meaning
- Grounding Multimodal Large Language Models in Actions (NeurIPS 2024 · similarity 0.88)
- Efficient Skill Grounding via Code Refactoring with Small Language Models (ICML 2026 · similarity 0.88)
- Ground-Compose-Reinforce: Grounding Language in Agentic Behaviours using Limited Data (NeurIPS 2025 · similarity 0.88)
- Grounding Language Plans in Demonstrations Through Counterfactual Perturbations (ICLR 2024 · similarity 0.88)
- EmbodiedBench: Comprehensive Benchmarking Multi-modal Large Language Models for Vision-Driven Embodied Agents (ICML 2025 · similarity 0.87)
- RoboDreamer: Learning Compositional World Models for Robot Imagination (ICML 2024 · 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.
