NEAT-NC: NEAT guided Navigation Cells for Robot Path Planning

A team of researchers has published a paper on NEAT-NC (NEAT guided Navigation Cells), a novel algorithm designed to revolutionize robot path planning. The work, led by Hibatallah Meliani, Khadija Slimani, and Samira Khoulji, is set to appear at the Genetic and Evolutionary Computation Conference (GECCO '26). The core innovation lies in its bio-inspired approach, directly modeling the brain's method of spatial navigation. Instead of traditional sensor inputs, the algorithm uses simulated versions of biological cells—like place cells, grid cells, and head direction cells—to create an internal cognitive map of an environment. These 'navigation cells' serve as the input to a neural network whose structure is evolved using the NEAT algorithm, which is known for optimizing both the weights and the topology of neural networks.

This fusion of evolutionary computation and neuroscience principles aims to solve a critical challenge: enabling autonomous agents to plan paths in unpredictable, dynamic settings. The researchers evaluated NEAT-NC across various static and dynamic scenarios, demonstrating its superior adaptability compared to standard NEAT. The evolved recurrent neural networks effectively represent a functional analog to the hippocampus, allowing for more robust and efficient route planning as environmental conditions change. The study concludes that this approach is particularly well-suited for applications requiring real-time decision-making, such as autonomous robotics and non-player character (NPC) AI in complex video games, marking a significant step toward more lifelike and resilient autonomous systems.