Prune-able fuzzy ART neural architecture for robot map learning and navigation in dynamic environments

Area of Science:

• Robotics
• Artificial Intelligence
• Neural Networks

Background:

• Mobile robots require robust mapping capabilities for navigation in unknown environments.
• Existing methods, like fuzzy ART neural architecture (FARTNA), have limitations in dynamic environments.
• Adapting to changing environments is crucial for autonomous navigation.

Purpose of the Study:

• To introduce a new online method for mobile robot map learning in unknown dynamic worlds.
• To extend the FARTNA with novel mechanisms for dynamic adaptation.
• To integrate object removal perception and shortest path planning for real-time navigation.

Main Methods:

• Introduction of the Prune-able fuzzy adaptive resonance theory neural architecture (PAFARTNA).
• Formulation and demonstration of relevant PAFARTNA properties.
• Integration of object removal perception and PAFARTNA into a navigation architecture.
• Real-time shortest path planning over a global world model.

Main Results:

• The PAFARTNA demonstrates dynamic adaptation capabilities.
• The integrated navigation architecture allows mobile robots to navigate changing worlds.
• A degree of optimality is maintained through real-time shortest path planning.
• Experimental results with a Nomad 200 robot validate the proposed methods.

Conclusions:

• The proposed PAFARTNA and integrated navigation architecture are effective for mobile robot navigation in dynamic environments.
• The method allows for online map learning and adaptation to world changes.
• The system maintains optimal path planning in real-time, enhancing robot autonomy.