An Efficient Sampling-Based Algorithms Using Active Learning and Manifold Learning for Multiple Unmanned Aerial...

Area of Science:

• Robotics and Control Systems
• Artificial Intelligence and Machine Learning
• Operations Research

Background:

• Task allocation for multiple unmanned aerial vehicles (UAVs) is complex, especially under uncertainty.
• Existing methods often face computational challenges and accuracy limitations.
• Efficiently modeling the impact of uncertainty on task rewards is crucial for optimal UAV operations.

Purpose of the Study:

• To develop a sampling-based approximation algorithm for multiple UAV task allocation under uncertainty.
• To reduce computational load while enhancing algorithmic accuracy.
• To create a high-precision evaluation model for uncertainty's impact on task rewards.

Main Methods:

• Constructing Gaussian process regression models using uncertainty parameters and task reward samples.
• Iteratively refining the training set via active learning and manifold learning.
• Employing a hybrid sampling strategy combining manifold learning for sample screening and multi-point sampling for efficient training set acquisition.

Main Results:

• A sparse graph representation of sample distribution is achieved using manifold learning.
• Active learning with multi-point sampling efficiently generates the training set.
• The proposed hybrid sampling strategy effectively selects representative samples.
• Simulation analyses confirm the algorithm's ability to generate high-precision evaluation models.

Conclusions:

• The developed sampling-based algorithm offers an effective approach for UAV task allocation under uncertainty.
• The hybrid sampling strategy significantly improves the efficiency and accuracy of model training.
• This method provides a robust tool for understanding and managing uncertainty in multi-UAV systems.