Football robot ball-carrying strategy selection method based on reinforcement learning

Abstract

The invention relates to a football robot ball-carrying strategy selection method based on reinforcement learning, a football field scene environment of a football robot is established on the basis ofa RoboCup simulation platform, and the method comprises the following steps: constructing a football robot ball target position model; decomposing the soccer robot ball target position model into a plurality of independent agents, obtaining a plurality of independent learners sharing the same state space and having different speed spaces, and setting a reward function for each independent learner; for each independent learner, constructing a reinforcement learning model based on SARSA (lambda), and carrying out approximate processing on the action value by adopting an RBF network; and training each independent learner, synchronously converging the learners by adopting a frequency adjustment learning method, obtaining a complete model, and completing ball-carrying strategy selection. Compared with the prior art, the invention has the advantages that the ball carrying process of the robot is quicker, the ball carrying is more controllable, the convergence is improved and the like.

Description

technical field [0001] The invention relates to the technical field of soccer robot sports, in particular to a method for selecting a football robot dribbling strategy based on reinforcement learning. Background technique [0002] In the RoboCup standard platform group competition, dribbling is a complex behavior during which the robot player tries to maneuver the ball in a very controlled manner while moving towards the desired goal. For a biped robot, the interaction between the ball, the robot, and the ground needs to be considered to obtain the velocities in the forward, lateral, and rotational directions, which makes this task highly dynamic and non-linear. Therefore, in the current technical solution, the dribbling behavior mostly controls the ball through a fixed speed and a fixed angle. What is the relationship between them? The robot always adjusts to the three-point line with the given angular speed and lateral movement speed, and then takes the ball to the target...

AI Technical Summary

Problems solved by technology

[0003] The purpose of the present invention is exactly to provide a kind of soccer robot dribbling strategy selection method based on reinforcement learning in order to overcome the defective that above-mentioned prior art exists, and this method is passed through reinforcement Learning is introduced into the selection of football robot dribbling strategy, and the three-dimensional movement is decomposed into three independent learners, and then the frequency adjustment learning method is used to make the three agents converge synchronously, which can not only dynamically adjust the speed, but also avoid reinforcement learning The problem of too high dimensionality is difficult to converge

Images