Learning sensor-based navigation of a real mobile robot in unknown worlds.

Area of Science:

• Robotics
• Artificial Intelligence
• Machine Learning

Background:

• Autonomous mobile robot navigation in unknown indoor environments presents significant challenges.
• Existing methods often involve inefficient exploration strategies and complex world modeling.

Purpose of the Study:

• To develop an improved, goal-oriented navigation method for autonomous mobile robots in unknown indoor environments.
• To enhance efficiency and reduce exploration costs during navigation.

Main Methods:

• Application of the parti-game multiresolution learning approach for cooperative world model construction.
• Integration of a fuzzy ART neural architecture for on-line map building from sensor data.
• Introduction of a predictive on-line trajectory filtering method for efficient path planning.

Main Results:

• The integrated approach allows for more efficient use of sensor information.
• Predictive trajectory filtering enables the robot to explore only successful trajectories, reducing exploration costs.
• Experiments with a Nomad 200 mobile robot demonstrated the effectiveness of the proposed navigation method.

Conclusions:

• The developed method offers a more efficient and cost-effective solution for autonomous robot navigation.
• The combination of parti-game learning, fuzzy ART, and predictive filtering significantly improves goal-oriented navigation.