Full-coverage task allocation method for dynamic noise environment

Abstract

The invention discloses a full-coverage task allocation method for a dynamic noise environment. The method comprises: firstly, constructing a multi-robot environment composed of robot nodes and a central control node, wherein each robot node is provided with a sensing information acquisition module, an input state estimation module, a position relation judgment module, an interaction weight estimation module and an action estimation module, a central control node is provided with an environment state monitoring module, an experience acquisition module and a network updating module, the perception information acquisition module acquires a local view and inputs the same into the input state estimation module to acquire an input state estimation vector and a numbered one-hot coding vector, the position relationship judgment module calculates an adjacent characteristic matrix set, an interaction weight estimation module calculates an adjacent weighted vector, and the action estimation module selects an action with the maximum estimation value as a decision, and the central control node optimizes the network in each module by adopting a reinforcement learning method. According to the method, execution and optimization are carried out simultaneously, task allocation is accurate and efficient, and the next task execution time of the robot is shorter.

Description

technical field [0001] The invention relates to the field of intelligent robot systems and multi-agent reinforcement learning technology, in particular to a multi-robot full-coverage task assignment method based on multi-agent reinforcement learning, which can be used in a dynamic environment with noise. Background technique [0002] Due to its mobility, mobile robots can replace humans to perform tasks such as exploration, detection and operation in various complex or dangerous environments. After years of research and development, mobile robots have gradually become practical. In traditional fields such as manufacturing, logistics, and service industries, there have been many examples of using mobile robots to improve production efficiency or replace manual operations. In some projects with great strategic significance to the national economy, society, national defense and other fields, the demand for mobile robots is becoming increasingly obvious. With the continuous exp...

AI Technical Summary

Problems solved by technology

2) When performing full-coverage task allocation tasks, there are often a lot of noise interference in the environment. For example, harsh environmental conditions may cause errors in the environmental perception information of a single robot, or hackers may control some robots to make their history Track record information is wrong

This requires the robot to learn to pay attention to the messages sent by the robots whose historical trajectories are around the robot's current position although the current distance is relatively far away. This requires an attention mechanism that accurately measures the interaction weight between each pair of robots. Most of the multi-agent reinforcement learning algorithms can only realize short-distance message attention weight calculation, and cannot balance the historical trajectory notification problem and the local target area conflict problem.

For example, if robot 1 and robot 3 are relatively close to each other, there is a potential conflict of goals and they should interact closely; similarly, robot 1 is about to explore the explored area of ​​robot 0, and the historical trajectory of robot 0 needs to be passed through message transmission. The form can tell robot 1 that although the two are far away at this time, the two should also interact closely to avoid repeated exploration, and the existing multi-agent reinforcement learning algorithm is difficult to balance the weight of these two parts of information

Images