Feature Importance Evaluation Method for Multi-Agent Deep Reinforcement Learning in Advanced Robotics Task Allocation

Abstract

The need to tackle intelligent tasks using advanced robotics multi-agent systems (MAS) actualize the use of artificial neural networks (ANNs) and multi-agent deep reinforcement learning technology. The article aims to solve the problem of exponential growth of ANN complexity with an increase in the number of agents in the MAS. To solve this problem, we propose an evaluation method for input data features importance. This method allows to optimize the input data feature set to reduce the computational complexity of the ANN inference while providing the same level of performance.