A Multi-Agent Action Prediction Algorithm Incorporating Decision Trees and Residual Networks
- Resource Type
- Conference
- Authors
- Yu, Yansong; Fang, Baofu
- Source
- 2023 4th International Conference on Big Data, Artificial Intelligence and Internet of Things Engineering (ICBAIE) Big Data, Artificial Intelligence and Internet of Things Engineering (ICBAIE), 2023 4th International Conference on. :125-131 Aug, 2023
- Subject
- Communication, Networking and Broadcast Technologies
Computing and Processing
Robotics and Control Systems
Signal Processing and Analysis
Neural networks
Fitting
Predictive models
Prediction algorithms
Market research
Real-time systems
Decision trees
AI
Deep Learning
multi-agent
- Language
The strategies of traditional multi-agent are difficult to understand from the outside through human observation due to the large joint state space and complex action behavior. Therefore, this paper investigates the utilization of expert log data in multi-agent and proposes a method for fitting strategies with residual networks based on expert datasets. The method can predict the agent’s forthcoming action and action parameters from the current state, so that the problem of predicting the opponent’s action in real time can be solved by using the expert data obtained from a large number of simulation environments. Results obtained from the RoboCup2D soccer simulation environment demonstrate that the algorithm effectively predicts agent actions and fits strategy models for different teams.