Sequential Localizing and Mapping: A Navigation Strategy via Enhanced Subsumption Architecture.

Area of Science:

• Robotics
• Artificial Intelligence
• Computer Vision

Background:

• Social robots require efficient navigation in domestic environments.
• Limited sensor capabilities pose challenges for robot autonomy.
• Integrating reactive and knowledge-based systems can enhance robot navigation.

Purpose of the Study:

• To develop and evaluate a navigation strategy for social robots with limited sensors.
• To integrate a reactive design with a learning knowledge system.
• To enable autonomous navigation in home environments.

Main Methods:

• A subsumption architecture combined with a learning knowledge system.
• Modules for doorway detection and room localization using convolutional neural networks (CNN).
• Obstacle avoidance via reinforcement learning and doorway passage using Canny edge detection.
• Creation of a Directional Semantic Topological Map (DST-Map).

Main Results:

• The proposed system successfully integrates various navigation modules.
• Convolutional neural networks (CNN) enabled effective doorway detection and room localization.
• Reinforcement learning facilitated efficient obstacle avoidance.
• The Directional Semantic Topological Map (DST-Map) provided an abstract representation for navigation.

Conclusions:

• The developed navigation strategy is effective for social robots with limited sensors in home environments.
• The hybrid approach combining reactive control and a learning knowledge system enhances robot navigation capabilities.
• The system's performance was validated in a virtual environment using the Webots simulator.