Robot soccer confrontation decision-making technology based on MOGM: Multi-objective game model

Abstract

Research on robot soccer confrontation decision-making technology (CDMT) has become a hot spot of the current artificial intelligence and robotics. But the current robot soccer CDMT has the defects of relying on static competition information and lacking the global consciousness. In this paper we propose a novel CDMT based on multi-objective game model (MOGM). MOGM is very suitable to multiple-robot competition. This technology establishes corresponding local information action-based game for every involved soccer robot. Then a global information strategy-based game by the linear weighted sum of the individual robots' local information games is evaluated. Thereby, the proposed method generates the global optimal strategy that meets the requirement of competition situation. The simulation results in Federation International Robot Soccer Association (FIRA) Standard Platform League show that this approach is efficient and reliable.

Cooperation Mode of Soccer Robot Game Based on Improved SARSA Algorithm

“…For example, An et al [11] from the Electronics Technology Laboratory in Japan focus on multi-intelligence collaborative systems in robotic football. Haobin et al [12] from Carnegie Mellon University in the USA have carried out work on the collaboration of intelligence using reinforcement learning, neural networks, and other methods.…”

“…In this context, action a is the action with the maximum value of Q among the actions considered in a given state s, which was considered to be n = 0:01 in the study by Haobin et al [12]. Thus, by using heuristics in the action selection method, learning is faster if the developer's domain knowledge is correct and the conditions are applied properly and slower if the conditions are applied improperly.…”

“…In the case of team SARSA, the decision is made to compare whether a goal was scored, the time it took to score, and the number of time the defence took more than 100 steps. In the case of team SARSA, the learning rate a = 1:25 is the same as in the case of [12]. Also, in team proposal B, the parameters of the heuristic were set to =0.5 and n = 0:01, and the results of the implementation were compared.…”

An adaptive cooperation with reinforcement learning for robot soccer games

“…19 Previous works using centralized optimization faced a lot of computational costs while they got an opportunity for developing an efficient collaboration. 20 So, it is often feasible that depends on decentralized optimization schemes.…”