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

TL;DR

NEAT-NC enhances NEAT with navigation cells for dynamic environment path planning.

cs.RO 🔴 Advanced 2026-04-16 36 views
Hibatallah Meliani Khadija Slimani Samira Khoulji
path planning neuroevolution dynamic environments robot navigation bio-inspired

Key Findings

Methodology

The paper introduces NEAT-NC, an algorithm that combines biological navigation cells with neuroevolution techniques. This method uses navigation cells such as place cells, border cells, head direction cells, and speed cells as inputs to evolve recurrent neural networks representing the hippocampus. This approach enables NEAT-NC to perform real-time path planning in dynamic environments, demonstrating adaptability in complex settings.

Key Results

  • Result 1: In static and dynamic environments, NEAT-NC achieved success rates of 93.33% and 100%, respectively, significantly outperforming Vanilla NEAT and DRL, which had success rates of 66.33% and 47%.
  • Result 2: NEAT-NC also excelled in path length and execution time, with an average path length of 1900.63 and execution time of 176.20 seconds, outperforming other algorithms.
  • Result 3: The Kruskal-Wallis and Dunn tests confirmed that NEAT-NC significantly outperformed other algorithms in path planning performance, with p-values less than 0.05.

Significance

This study significantly enhances NEAT's performance in dynamic path planning by introducing the concept of biological navigation cells. The method is not only academically significant but also offers new solutions for real-time path planning in robotics and games. By simulating the brain's spatial cognition capabilities, NEAT-NC demonstrates the potential of bio-inspired algorithms in complex environments.

Technical Contribution

NEAT-NC provides a novel path planning method by integrating biological navigation cells and recurrent neural networks. Compared to existing SOTA methods, this algorithm shows significant improvements in adaptability and efficiency in dynamic environments. Additionally, NEAT-NC's flexible fitness function design effectively guides evolutionary search towards more efficient navigation behaviors.

Novelty

NEAT-NC is the first to integrate the concept of biological navigation cells into the NEAT algorithm, forming a novel path planning framework. Compared to previous studies, this method innovates not only in algorithm structure but also in the application of biological theories.

Limitations

  • Limitation 1: NEAT-NC may incur high computational costs when handling very complex dynamic environments, affecting real-time performance.
  • Limitation 2: The algorithm heavily relies on sensor accuracy, and sensor errors may impact navigation effectiveness.
  • Limitation 3: In some extreme cases, the algorithm may require more training time to achieve optimal performance.

Future Work

Future research directions include further optimizing NEAT-NC's computational efficiency, exploring more applications of biological navigation cells, and integrating navigation with manipulation to form a fully interactive agent operating in a 3D environment.

AI Executive Summary

Path planning is a critical issue in modern robotics, especially in dynamic environments. Traditional methods such as genetic algorithms and particle swarm optimization, while effective, often perform poorly in complex environments. NEAT-NC offers a new solution by introducing the concept of biological navigation cells. This method uses place cells, border cells, head direction cells, and speed cells as inputs to evolve recurrent neural networks representing the hippocampus. This bio-inspired design allows NEAT-NC to achieve efficient path planning in dynamic environments.

The core technology of NEAT-NC includes using recurrent neural networks as spatial memory, combined with information from biological navigation cells to guide agents in complex environments. Experimental results show that NEAT-NC performs excellently in multiple test environments, with success rates and path efficiency surpassing traditional algorithms. This indicates the method's great potential in real-time path planning.

By simulating the brain's spatial cognition capabilities, NEAT-NC demonstrates the potential of bio-inspired algorithms in complex environments. This research is not only academically significant but also offers new solutions for real-time path planning in robotics and games.

Despite NEAT-NC's outstanding performance in many aspects, it may incur high computational costs when handling very complex dynamic environments, affecting real-time performance. Additionally, the algorithm heavily relies on sensor accuracy, and sensor errors may impact navigation effectiveness. Future research will focus on optimizing the algorithm's computational efficiency and exploring more applications of biological navigation cells.

In conclusion, NEAT-NC provides an innovative solution to the path planning problem by combining biological navigation cells and neuroevolution techniques. Future research will continue to explore more possibilities in this field, promoting the application of agents in complex environments.

Deep Analysis

Background

Path planning is a critical issue in robotics, especially in dynamic environments. Traditional methods such as genetic algorithms, simulated annealing, and particle swarm optimization, while effective, often perform poorly in complex environments. In recent years, with the rapid development of artificial intelligence and machine learning technologies, researchers have begun exploring bio-inspired algorithms such as NeuroEvolution of Augmenting Topologies (NEAT). NEAT optimizes both the weights and structure of neural networks, showing excellent performance in dynamic and unpredictable environments. However, how to further enhance NEAT's performance in dynamic path planning remains a challenge.

Core Problem

Path planning in dynamic environments is a complex problem that involves finding a feasible obstacle-free path in optimal time. Traditional methods perform well with static obstacles but often struggle to adapt in dynamic environments. This is because obstacles in dynamic environments constantly change, requiring algorithms to be more adaptive and flexible. Additionally, effectively utilizing sensor data for environmental perception and path planning is a significant challenge.

Innovation

NEAT-NC introduces a new path planning method by incorporating the concept of biological navigation cells. • This method uses place cells, border cells, head direction cells, and speed cells as inputs to evolve recurrent neural networks representing the hippocampus. • This bio-inspired design allows NEAT-NC to achieve efficient path planning in dynamic environments. • Compared to traditional NEAT algorithms, NEAT-NC innovates in algorithm structure by integrating information from biological navigation cells, significantly enhancing the algorithm's adaptability and efficiency.

Methodology

The core methodology of NEAT-NC includes the following steps: • Use navigation cells (such as place cells, border cells, head direction cells, and speed cells) as inputs, simulating spatial cognition capabilities in biological systems. • Recurrent Neural Networks (RNN) serve as spatial memory, combining information from navigation cells to guide agents in complex environments. • A flexible fitness function is designed to effectively guide evolutionary search towards more efficient navigation behaviors. • The algorithm evolves optimal neural network structures through genetic operations such as selection, crossover, and mutation.

Experiments

The experimental design includes testing NEAT-NC's performance in three different environments, including static and dynamic obstacles. • Dynamic obstacles move horizontally or vertically at constant speeds within predefined ranges, adding dynamic elements to the environment. • To ensure fair comparison, Vanilla NEAT and PPO were used as baseline algorithms in the experiments. • The experimental results were evaluated through metrics such as success rate, path length, execution time, and average fitness value.

Results

The experimental results show that NEAT-NC performs excellently in multiple test environments. • In static and dynamic environments, NEAT-NC achieved success rates of 93.33% and 100%, respectively, significantly outperforming Vanilla NEAT and DRL. • NEAT-NC also excelled in path length and execution time, with an average path length of 1900.63 and execution time of 176.20 seconds, outperforming other algorithms. • The Kruskal-Wallis and Dunn tests confirmed that NEAT-NC significantly outperformed other algorithms in path planning performance, with p-values less than 0.05.

Applications

NEAT-NC has broad application potential in real-time path planning for robotics and games. • In robotics, NEAT-NC can be used for path planning in autonomous vehicles, drones, and other autonomous mobile devices. • In games, NEAT-NC can be used for intelligent navigation of characters, enhancing game interactivity and challenge. • These applications require high-precision sensor data and powerful computing capabilities to ensure the algorithm's real-time performance and accuracy.

Limitations & Outlook

Despite NEAT-NC's outstanding performance in many aspects, it may incur high computational costs when handling very complex dynamic environments, affecting real-time performance. • The algorithm heavily relies on sensor accuracy, and sensor errors may impact navigation effectiveness. • In some extreme cases, the algorithm may require more training time to achieve optimal performance. Future research will focus on optimizing the algorithm's computational efficiency and exploring more applications of biological navigation cells.

Plain Language Accessible to non-experts

Imagine you're in a maze where the walls can move, and you need to find a safe path to the end. NEAT-NC is like your brain, helping you sense the environment and adjust your path based on the moving walls. It uses something called 'navigation cells,' which are like your eyes and ears, helping you sense obstacles and the target location. Then, NEAT-NC acts like a smart guide, telling you which direction to go and how fast. Even if the maze layout keeps changing, it helps you find the quickest path. This method is useful not only in robot navigation but also in games, making characters move more intelligently.

ELI14 Explained like you're 14

Hey there! Imagine you're playing a super cool maze game where the walls can move! You need to find a safe path to the end. NEAT-NC is like your game assistant, helping you sense the environment and adjust your path based on the moving walls. It uses something called 'navigation cells,' which are like your eyes and ears, helping you sense obstacles and the target location. Then, NEAT-NC acts like a smart guide, telling you which direction to go and how fast. Even if the maze layout keeps changing, it helps you find the quickest path. This method is useful not only in robot navigation but also in games, making characters move more intelligently. Isn't that cool?

Glossary

NEAT (NeuroEvolution of Augmenting Topologies)

An evolutionary algorithm that optimizes both the weights and structure of neural networks, suitable for dynamic and unpredictable environments.

In this paper, NEAT is used for path planning by evolving neural network structures to adapt to dynamic environments.

Navigation Cells

Simulate spatial cognition capabilities in biological systems, including place cells, border cells, head direction cells, and speed cells.

In NEAT-NC, navigation cells serve as inputs to help agents perceive the environment.

Recurrent Neural Network (RNN)

A neural network structure capable of processing sequential data with memory capabilities.

In NEAT-NC, RNN is used to simulate spatial memory, combining information from navigation cells for path planning.

Place Cells

Cells in biological systems that activate when an organism is in a specific location, helping form spatial memory.

In NEAT-NC, place cells are used to perceive the target location.

Border Cells

Cells in biological systems that activate when an organism approaches obstacles or boundaries.

In NEAT-NC, border cells are used to perceive the position of obstacles.

Head Direction Cells

Cells in biological systems that activate when an organism's head is oriented in a specific direction.

In NEAT-NC, head direction cells are used to perceive the target direction.

Speed Cells

Cells in biological systems where activation rate depends on the individual's running speed.

In NEAT-NC, speed cells are used to perceive the agent's speed.

Fitness Function

Used to evaluate an individual's performance in the evolutionary process, guiding evolutionary search.

In NEAT-NC, the fitness function is flexibly designed to effectively guide evolutionary search.

Dynamic Environment

An environment where obstacles constantly change, requiring algorithms to be more adaptive and flexible.

NEAT-NC is designed specifically for path planning in dynamic environments.

Deep Reinforcement Learning (DRL)

Combines deep learning and reinforcement learning methods to train agents to make decisions in complex environments.

In the experiments, PPO is used as a baseline algorithm for deep reinforcement learning.

Open Questions Unanswered questions from this research

  • 1 How to enhance NEAT-NC's performance in ultra-complex dynamic environments without increasing computational costs? Existing methods face bottlenecks in computational efficiency, requiring more efficient algorithm designs.
  • 2 How to reduce reliance on sensor accuracy? Existing methods have high requirements for sensor data accuracy, and sensor errors may impact navigation effectiveness.
  • 3 How to integrate navigation with manipulation to form a fully interactive agent operating in a 3D environment? Current research mainly focuses on path planning in 2D environments.
  • 4 How to further optimize NEAT-NC's fitness function design to perform excellently in more scenarios? Existing fitness functions may need adjustment in some extreme cases.
  • 5 How to improve NEAT-NC's adaptability without increasing training time? Existing methods may require longer training times in some cases to achieve optimal performance.

Applications

Immediate Applications

Autonomous Vehicles

NEAT-NC can be used for path planning in autonomous vehicles, helping them navigate safely and efficiently in dynamic environments.

Drone Navigation

In drone navigation, NEAT-NC can help drones find optimal paths in complex environments, avoiding obstacles.

Game Character Navigation

In games, NEAT-NC can be used for intelligent navigation of characters, enhancing game interactivity and challenge.

Long-term Vision

Smart City Traffic Management

NEAT-NC can be used in smart city traffic management systems to optimize traffic flow and improve urban traffic efficiency and safety.

Fully Automated Warehouse Management

In fully automated warehouses, NEAT-NC can be used for robot path planning, improving warehouse operational efficiency and accuracy.

Abstract

To navigate a space, the brain makes an internal representation of the environment using different cells such as place cells, grid cells, head direction cells, border cells, and speed cells. All these cells, along with sensory inputs, enable an organism to explore the space around it. Inspired by these biological principles, we developed NEATNC, a Neuro-Evolution of Augmenting Topology guided Navigation Cells. The goal of the paper is to improve NEAT algorithm performance in path planning in dynamic environments using spatial cognitive cells. This approach uses navigation cells as inputs and evolves recurrent neural networks, representing the hippocampus part of the brain. The performance of the proposed algorithm is evaluated in different static and dynamic scenarios. This study highlights NEAT's adaptability to complex and different environments, showcasing the utility of biological theories. This suggests that our approach is well-suited for real-time dynamic path planning for robotics and games.

cs.RO cs.AI cs.NE

References (20)

Hybrid path planning algorithm for robots based on modified golden jackal optimization method and dynamic window method

Yuchao Wang, Kelin Tong, Chunhai Fu et al.

2025 12 citations

Brain-like path planning algorithm based on spiking neural network

Jilun Zhang, Ying Liu

2025 2 citations

TempGA: A Temperature-Inspired Adaptive Genetic Algorithm for Solving 7DOF Inverse Kinematics Problems

Hibatallah Meliani, Khadija Slimani, Samira Khoulji

2025 1 citations

Grid cells in pre- and parasubiculum

Charlotte N. Boccara, F. Sargolini, Veslemøy Hult Thoresen et al.

2010 503 citations

Neuroevolutionary multi-objective approaches to trajectory prediction in autonomous vehicles

Fergal Stapleton, E. López, Ganesh Sistu et al.

2022 11 citations View Analysis →

The hippocampus as a spatial map. Preliminary evidence from unit activity in the freely-moving rat.

J. O’Keefe, J. Dostrovsky

1971 6063 citations

Path Planning of Mobile Robot Based on Dual-Layer Fuzzy Control and Improved Genetic Algorithm

Yangxin Teng, Tingping Feng, Changlin Song et al.

2025 7 citations

Towards a Predictive Bio-Inspired Navigation Model

S. L. Gay, Kévin Le Run, E. Pissaloux et al.

2021 14 citations

Dendritic mechanisms of hippocampal place field formation.

M. Sheffield, D. Dombeck

2019 50 citations

Direct recordings of grid-like neuronal activity in human spatial navigation

2013 406 citations

A Spatial Cognitive Cells Inspired Goal-directed Navigation Model

Lingfang Hu, K. Hao, Xin Cai et al.

2019 3 citations

Optimal path planning approach based on Q-learning algorithm for mobile robots

Abderraouf Maoudj, A. Hentout

2020 126 citations

A Bio-Inspired Goal-Directed Cognitive Map Model for Robot Navigation and Exploration

Matthew Hicks, Tingjun Lei, Chaomin Luo et al.

2025 9 citations

Hippocampal place cells have goal-oriented vector fields during navigation

J. Ormond, John O’Keefe

2022 91 citations

NEAT-based 3D path planning for mobile robotic arms in NDT with offline inverse kinematics validation

Mengyuan Zhang, Qingping Yang, M. Sutcliffe et al.

2025 1 citations

Real-Time and Energy-Aware UAV Routing: A Scalable DAR Approach for Future 6G Systems

K. Slimani, Samira Khoulji, Hamed Taherdoost et al.

2025 1 citations

Hippocampus-independent phase precession in entorhinal grid cells

T. Hafting, M. Fyhn, Tora Bonnevie et al.

2008 492 citations

Mobile Robot Path Planning using Deep Deterministic Policy Gradient with Differential Gaming (DDPG-DG) exploration

Shripad Deshpande, H. R, Babul Salam Ksm Kader Ibrahim et al.

2024 14 citations

The head direction signal: origins and sensory-motor integration.

J. Taube

2007 889 citations

Place Cells: The Brain Cells That Help us Navigate the World

J. O’Keefe

2023 1 citations