| Multi-mobile robot systems are widely recognized for their high efficiency and robustness in real-world applications.Path planning and behavioral decision-making technologies,as the "brain" organization layers of the robots’ navigation process,express the intelligence level of robots.This paper takes simulation software and mobile robots as the experimental platforms and studies the path planning and behavioral decision-making methods of multi-mobile robots for different cooperative tasks.The main contents are as follows:(1)Aiming at the problems of conflicting targets and unbalanced task load in multi-robot cleaning tasks,a multi-target-oriented cooperative coverage path planning method for cleaning robots is proposed.First,the environment map is obtained and preprocessed.Further,the coverage path planning is carried out based on the multi-target fuzzy comprehensive analysis method to achieve the high coverage rate and low repetition rate of multi-robot cleaning paths.Then,the variance of the task load is designed as a fitness function,and the genetic algorithm is used to divide the coverage path.The division result is the cleaning path of each robot.This realizes the simultaneous cleaning for multiple robots.Finally,based on the robot operating system,a cooperative coverage path planning system for cleaning robots is designed,which is successfully applied in the real environment.(2)With the increase in the number of environmental participants,it becomes difficult for robots to process a large amount of environmental information,and even leads to failure in decision making.Thus,a multirobot cooperative hunting decision-making and path planning method based on hybrid attention-oriented experience replay is proposed.A hybrid attention module is designed to extract key information in the environmental state and joint actions,respectively.This module,embedded in multi-agent deep deterministic policy gradient,improves the information processing capabilities of multiple robots.At the same time,a hybrid attention-oriented priority experience replay method is designed to improve the utilization of multi-robot experience samples,thereby speeding up the convergence.The method exhibits excellent convergence performance and scalability in the predator-prey environment and is successfully applied to the real-world multi-robot hunting decision-making and path planning task.(3)Aiming at the problem that discrete actions and continuous actions have a serious coupled relationship in behavioral decision-making based on parameterized action space,a cooperative offensive decision-making method for soccer robots based on bi-channel Q-value evaluation is put forward.The reward function is designed at the angle level to guide the soccer robot to shoot at the appropriate goal point,which can accelerate the training process.In addition,a critic network with bi-channel Q-value evaluation is proposed.The state-discrete action value and the state-hybrid action value are estimated separately,which improves the robustness of network learning.Moreover,according to the staged characteristics of the cooperative offensive decision-making training process of soccer robots,a discrete action weight parameter is designed to adjust adaptively,so that the network can learn discrete actions and continuous actions with emphasis in the training process.The proposed method achieves a high goal rate in the half field offense environment and has strong robustness and scalability.In summary,this paper proposes several solutions for multi-robot cooperative path planning and decision-making under cleaning tasks,hunting tasks,and soccer competitions,and conducts relevant simulations and experiments.The result shows that the proposed method can successfully achieve multi-robot cooperative tasks and has good scalability and stability. |
