Research map / Large language models / AI and society
🏘️ ProcTHOR: Large-Scale Embodied AI Using Procedural Generation
Matt Deitke, Eli VanderBilt, Alvaro Herrasti, Luca Weihs, Kiana Ehsani, Jordi Salvador, Winson Han, Eric Kolve, Aniruddha Kembhavi, Roozbeh Mottaghi
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 |
|---|---|---|---|
| Pre-train on ProcTHOR and evaluate zero-shot instead of using the RoboTHOR-trained baseline | ObjectNav agent · RoboTHOR Challenge | SPL: 0.200 -> 0.237 | Improved |
| Fine-tune the ProcTHOR-pretrained model on benchmark training data | ObjectNav agent · RoboTHOR Challenge | SPL: 0.237 -> 0.288 | Improved |
| Use ProcTHOR pre-training and fine-tune on the benchmark | ObjectNav agent · Habitat Challenge 2022 (HM3D-Semantics) | SPL: 0.055 -> 0.270; Large: 0.077 -> 0.318 | Mixed |
| Pre-train on ProcTHOR and evaluate zero-shot instead of using the AI2-iTHOR-trained baseline | ObjectNav agent · AI2-iTHOR ObjectNav | SPL: 0.516 -> 0.644 | Improved |
| Fine-tune the ProcTHOR-pretrained model on benchmark training data | ObjectNav agent · AI2-iTHOR ObjectNav | SPL: 0.644 -> 0.621 | Hurt |
| Train on 10,000 generated houses instead of fewer houses (ablation) | ProcTHOR ObjectNav agent · ProcTHOR ObjectNav | SPL: 0.077 -> 0.185 | Improved |
| Train on 10,000 generated houses instead of fewer houses (ablation) | ObjectNav agent · RoboTHOR Test (0-shot) | SPL: 0.040 -> 0.210 | Improved |
| Train on 10,000 generated houses instead of fewer houses (ablation) | ObjectNav agent · AI2-iTHOR Test (0-shot) | SPL: 0.249 -> 0.554 | Improved |
| Use a larger CLIP backbone and wider RNN (ablation) | ProcTHOR ObjectNav agent · Habitat Challenge 2022 (HM3D-Semantics) | SPL: 0.055 -> 0.077 | Improved |
| Fine-tune the ProcTHOR-pretrained model on benchmark training data | Rearrangement agent · 1-phase Rearrangement Challenge 2022 | % Fixed Strict: 0.156 -> 0.245 | Improved |
| Pre-train on ProcTHOR and evaluate zero-shot instead of using the iTHOR-SimpleConv baseline | ArmPointNav agent · ManipulaTHOR | Success: 29.2% -> 37.9% | Improved |
Closest papers by meaning
- EmbodiedBench: Comprehensive Benchmarking Multi-modal Large Language Models for Vision-Driven Embodied Agents (ICML 2025 · similarity 0.87)
- VirtualEnv: A Platform for Embodied AI Research (AAAI 2026 · similarity 0.87)
- DynScene: Scalable Generation of Dynamic Robotic Manipulation Scenes for Embodied AI (CVPR 2025 · similarity 0.86)
- EmbodiedScan: A Holistic Multi-Modal 3D Perception Suite Towards Embodied AI (CVPR 2024 · similarity 0.86)
- TPRU: Advancing Temporal and Procedural Understanding in Large Multimodal Models (ICLR 2026 · similarity 0.86)
- Holodeck: Language Guided Generation of 3D Embodied AI Environments (CVPR 2024 · similarity 0.86)
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.
