Learning probabilistic features for robotic navigation using laser sensors.

Area of Science:

• Robotics
• Artificial Intelligence
• Computer Vision

Background:

• Simultaneous Localization and Mapping (SLAM) is crucial for autonomous systems to map unknown environments and determine their position.
• Previous SLAM implementations faced high computational complexities, often O(Nlog(N)) to O(N^2), limiting their practical application.
• Developing efficient SLAM algorithms is essential for advancing robotic capabilities in real-world scenarios.

Purpose of the Study:

• To introduce a novel SLAM-based probabilistic robotic system capable of learning environmental features.
• To significantly reduce the computational complexity of SLAM algorithms.
• To enable low-cost robots to perform autonomous mapping and localization.

Main Methods:

• The system employs a probabilistic approach, utilizing a model to fuse sensor information via the Bayesian paradigm.
• Computational complexity is reduced to O(N) through an innovative data fusion technique.
• The robot learns essential features of its environment during a training phase.

Main Results:

• The developed SLAM system achieves a computational complexity of O(N), a substantial improvement over prior methods.
• Post-training, the robot can effectively identify and locate previously learned areas within the environment.
• A three-dimensional map is generated using a single laser sensor, demonstrating effective environmental perception.

Conclusions:

• The proposed SLAM system offers a computationally efficient solution for robotic mapping and localization.
• The system's ability to learn and recognize environmental features enhances its autonomy.
• The use of low-complexity algorithms makes the system suitable for implementation on resource-constrained, low-cost robots.