Efficient Clustering for Continuous Occupancy Mapping Using a Mixture of Gaussian Processes.

Area of Science:

• Robotics
• Artificial Intelligence
• Machine Learning

Background:

• Gaussian processes offer high accuracy for robotic occupancy mapping.
• High computational complexity of Gaussian processes limits large-scale applications.
• Clustering data is necessary but determining the number of clusters is challenging.

Purpose of the Study:

• To propose a novel, computationally efficient method for occupancy map building.
• To address the challenge of unknown cluster numbers in data clustering.
• To improve the speed and accuracy of Gaussian process-based mapping.

Main Methods:

• Employing a mixture of Gaussian processes for occupancy mapping.
• Utilizing Dirichlet processes for unsupervised clustering of data.
• Incorporating geometrical information (e.g., line features) to enhance clustering.
• Developing two efficient clustering methods: Dirichlet process-based and geometry-based.

Main Results:

• Dirichlet process-based clustering significantly speeds up Gaussian process training.
• Geometrical features further improve clustering accuracy in occupancy mapping.
• The proposed methods demonstrate significant benefits in simulation results.

Conclusions:

• The novel mixture of Gaussian processes with Dirichlet process and geometric clustering offers an efficient solution for large-scale robotic occupancy mapping.
• This approach overcomes the computational limitations of traditional Gaussian processes.
• The methods provide a robust and accurate way to build occupancy maps.