Intention Recognition Algorithm for Multi-agent Systems Based on High-order Fully Actuated System Approach

Intention recognition for multiple agents is an important problem in multi-agent systems (MASs), and is widely used in the field of autonomous driving, human-machine interaction and military. In order to improve the competitive ability in multi-agent confrontation, an intention recognition algorithm for multi-agent systems based on high-order fully actuated (HOFA) system approach is proposed. Due to the uncertainty of the closed-loop system of the agents, the HOFA system approach is introduced to generate a data set with more extensive features, and an algorithm for the data set establishment is proposed. To obtain the intention prediction results, an intention recognition model based on artificial neural networks is proposed. Structures of both convolutional neural networks and recurrent neural networks are introduced to process the time features and spatial features. The intention predictor is trained via the data set based on HOFA system approach and tested on the test set. The simulation results shows that the proposed predictor has a better performance for intention recognition problem.