Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Arbitrarily Scalable Environment Generators via Neural Cellular Automata
Yulun Zhang, Matthew Christopher Fontaine, Varun Bhatt, Stefanos Nikolaidis, Jiaoyang Li
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 |
|---|---|---|---|
| increase alpha from 0 to 5 in CMA-MAE (ablation) | CMA-MAE + NCA · Warehouse (even) size S_eval | success rate: 0% -> 90% | Improved |
| increase alpha from 0 to 5 in CMA-MAE (ablation) | CMA-MAE + NCA · Warehouse (uneven) size S_eval | success rate: 62% -> 84% | Improved |
| increase alpha from 1 to 5 in CMA-MAE (ablation) | CMA-MAE + NCA · Warehouse (uneven) size S_eval | success rate: 8% -> 84% | Improved |
| increase alpha from 0 to 5 (ablation) | CMA-MAE + NCA · Warehouse (uneven) size S_eval | throughput: 12.32 -> 12.03 | Mixed |
| vary alpha (0, 1, 5) (ablation) | CMA-MAE + NCA · Warehouse size S | throughput: 6.79, 6.73, 6.74 | No clear effect |
| NCA-generated vs DSAGE-optimized environments | CMA-MAE + NCA · Warehouse (even) size S | throughput: 6.35 -> 6.74 | Improved |
| NCA-generated vs DSAGE-optimized environments | CMA-MAE + NCA · Manufacturing size S | throughput: 5.61 -> 6.61 | Improved |
| NCA-generated vs DSAGE on scalability with more agents | CMA-MAE + NCA · Warehouse (uneven) | throughput vs agents | Hurt |
| NCA-generated vs human-designed environments | CMA-MAE + NCA · Manufacturing size S | throughput: 5.92 -> 6.82 | Improved |
| NCA-generated vs human-designed environments | CMA-MAE + NCA · Warehouse (even) size S_eval | success rate: 0% -> 90% | Improved |
| NCA-generated maze vs baseline maze with same walls and path lengths | ACCEL agent · Maze 66x66 | success rate: 22% -> 93% | Improved |
| scale NCA-generated maze from 18x18 to 66x66 | ACCEL agent · Maze | success rate: 90% -> 93% | No clear effect |
Closest papers by meaning
- Agent World Model: Infinity Synthetic Environments for Agentic Reinforcement Learning (ICML 2026 · similarity 0.85)
- Don't Just Fine-tune the Agent, Tune the Environment (ICLR 2026 · similarity 0.84)
- Agent2World: Learning to Generate Symbolic World Models via Adaptive Multi-Agent Feedback (NeurIPS 2026 · similarity 0.84)
- GenEnv: Difficulty-Aligned Co-Evolution Between LLM Agents and Environment Simulators (NeurIPS 2026 · similarity 0.84)
- SimWorld Studio: Automatic Environment Generation with Evolving Coding Agent for Embodied Agent Learning (NeurIPS 2026 · similarity 0.83)
- PARCO: Parallel AutoRegressive Models for Multi-Agent Combinatorial Optimization (NeurIPS 2025 · similarity 0.83)
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.
