Trajectory Modeling by Distributed Gaussian Processes in Multiagent Systems.

Area of Science:

• Robotics
• Artificial Intelligence
• Control Systems

Background:

• Multi-agent systems face challenges in modeling complex environments with uncertainties and noise.
• Data-driven methods like Gaussian processes are effective for characterizing model uncertainties.
• Existing methods often require central coordination, limiting scalability.

Purpose of the Study:

• To propose a distributed Gaussian process for robust trajectory modeling in multi-agent systems.
• To address model uncertainties and noise disturbances through local information exchange.
• To develop control strategies for both continuous-time and discrete-time system models.

Main Methods:

• A distributed Gaussian process framework utilizing local measurements and neighbor data exchange.
• Design of a control Lyapunov function for continuous-time model learning.
• Application of a distributed model predictive control approach for discrete-time model learning.
• Implementation of a Kullback-Leibler average consensus fusion algorithm for prediction fusion.

Main Results:

• The proposed distributed Gaussian process effectively characterizes system models with uncertainties.
• Successful trajectory tracking demonstrated through two illustrative examples.
• Cooperative estimation of a common Gaussian process function achieved through local interactions.

Conclusions:

• The distributed Gaussian process offers a robust and scalable solution for multi-agent trajectory modeling.
• The developed control strategies enable effective learning of both continuous-time and discrete-time models.
• The fusion algorithm ensures accurate aggregation of local predictions for enhanced system performance.