Visual topology navigation method based on reinforcement learning

Abstract

The invention discloses a visual topology navigation method based on reinforcement learning. The method comprises the steps of: building a topological map based on a reinforcement learning model and a sequence image, giving a current position and a target position of a mobile robot, and searching and matching the current position and the target position based on the topological map; after the current position and the target position are positioned, carrying out path planning, and selecting a shortest path from the current position to the target position by adopting a shortest path algorithm; predicting a sub-target node, and selecting one node from the nodes after the current node in the shortest path as the sub-target node to guide the mobile robot to reach the next sub-target node; and taking the current state image features and the state image features of the sub-target nodes as input of a reinforcement learning network, performing navigation control by the reinforcement learning network, and outputting an action signal for guiding the mobile robot to reach the sub-target nodes until navigation to the target position is finished.

Description

technical field [0001] The invention belongs to the field of visual navigation of mobile robots, in particular to a visual topology navigation method based on reinforcement learning. Background technique [0002] Autonomous navigation is the basic premise for mobile robots to realize environmental exploration, and it is also a hotspot of current research. Reinforcement learning has been considered as a promising technique for autonomous exploration due to its outstanding action planning ability. Reinforcement learning is a self-evolving type of machine learning that learns by interacting with the environment and learning through continuous trial and error, which is closer to realizing real artificial intelligence. However, because reinforcement learning is limited by its own planning ability, it is powerless for long-distance navigation, so planning algorithms are needed to guide and decompose large-scale navigation tasks into subtasks that reinforcement learning can comple...

Images