2D Spatial Reasoning with Adaptive Neural Cellular Automata
On the map: LLM reasoning → Spatial and embodied reasoning
Abstract (arXiv)
Many modern learning approaches are still struggling with spatial reasoning tasks, i.e. they lack the ability to utilize geometric information of perceived entities and their spatial relation to each other to solve problems. We introduce a novel Adaptive Neural Cellular Automata (aNCA) architecture which uses deformable convolutions to dynamically adapt the perceptive field and iteratively reason over 2D spatial relations on grid-like data structures (e.g. images). Empirical results on public benchmarks show state of the art comprehensible results with high generalization abilities for solving image based puzzles like Sudoku or finding the shortest path in a maze.
Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Spatial and embodied reasoning, one of 17 directions in LLM reasoning. First posted on arXiv 6 Oct 2026.
Builds on
No reference list yet.
Built on by
Not cited yet.
Similar, not cited
Close in wording, no citation either way
- 2026Reasoning with Neural Cellular Automata
- 2026Active Exploring like a Pigeon: Reinforcing Spatial Reasoning via Agentic Vision-Language Models
- 2026Relational Abstractions for Spatial Reasoning with Diffusion Models
- 2025Ascending the Infinite Ladder: Benchmarking Spatial Deformation Reasoning in Vision-Language Models
- 2025Bridging the Dynamic Perception Gap: Training-Free Draft Chain-of-Thought for Dynamic Multimodal Spatial Reasoning
- 2024SAT: Dynamic Spatial Aptitude Training for Multimodal Language Models
- 2025SPaRC: A Spatial Pathfinding Reasoning Challenge
- 2025ORIGAMISPACE: Benchmarking Multimodal LLMs in Multi-Step Spatial Reasoning with Mathematical Constraints