Mobile robot navigation method

Abstract

The invention discloses a mobile robot navigation method. An image acquisition device is arranged overhead in a navigation area. The method comprises the following steps: step one, collecting a mobilerobot environment image once before the robot motion; step two, acquiring environmental obstacle information according to the collected environment image; step three, establishing a binary raster mapbased on the obtained environmental obstacle information and marking a passable area and non-passage area on the raster map; step four, establishing a mobile robot running rule based on the raster map; step five, setting a total round number M and a shallow trial learning round number M1; to be specific, carrying out shallow trial learning based on a mobile robot running rule to obtain a preliminary Q table; and updating the Q table by reinforced learning based on an initial position p0 of the mobile robot; an step six, according to the updated Q table, acquiring an optimal motion strategy pai<*> of the mobile robot to obtain a motion path of the mobile robot. Therefore, the invalid exploration of the robot in the training process is reduced; the learning efficiency is high and the convergence is fast.

Description

technical field [0001] The invention relates to a reinforcement learning algorithm under the background of human prior knowledge to provide effective navigation for mobile robots. Background technique [0002] A key skill for mobile robots is the ability to efficiently navigate their environment, and reinforcement learning is widely used for path planning for mobile robots. However, this algorithm has a slow convergence rate and a large number of iterations. There is little research on how to effectively improve learning from an acquisition perspective in rule-based shallow trial strategies. In the biological world, animals rely on their own empirical knowledge when making path planning. Human nature has prior knowledge that helps people a lot in their voyages. We take the prior knowledge of human behavior and express it as shallow rules, and then apply the rule-based shallow reinforcement learning to the robot's navigation learning, effectively improving the learning eff...

AI Technical Summary

Problems solved by technology

However, this algorithm has a slow convergence rate and a large number of iterations

There is little research on how to effectively improve learning from an acquisition perspective in rule-based shallow trial strategies

Images