New framework iARCS generates controllable 3D scenes using agentic RL

Researchers have developed iARCS, a novel framework that uses iterative agentic reinforcement learning to generate controllable 3D scenes. This method adapts pre-trained scene generators to meet natural-language task requirements by first improving physical plausibility and then fine-tuning with iteratively refined reward programs generated by LLMs. Experiments demonstrate that iARCS enhances constraint fidelity for tasks like walkability and reachability, while also maintaining scene diversity and improving downstream training data. AI

IMPACT Enables more functional and task-specific synthetic data generation for computer vision and embodied AI.

RANK_REASON The cluster contains a research paper detailing a new framework for 3D scene generation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• alphaXiv
• arXiv
• CatalyzeX
• Connected Papers
• CORE Recommender
• DagsHub
• Gotit.pub
• Hugging Face
• iARCS
• Litmaps
• ScienceCast
• scite Smart Citations

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →