Robot indoor complex scene obstacle avoidance method based on monocular camera

Abstract

The invention discloses a robot indoor complex scene obstacle avoidance method based on a monocular camera, and belongs to the field of robot navigation and obstacle avoidance. The monocular obstacle avoidance navigation network is composed of an environment perception stage and a control decision stage, and specifically comprises a depth prediction module, a semantic mask module, a depth slicing module, a feature extraction guidance module, a reinforcement learning module and data enhancement. According to the network, a monocular RGB image serves as input, after a semantic depth map is obtained, dynamic minimum pooling operation is conducted to obtain pseudo laser data; then the pseudo laser data serve as state input of reinforcement learning, and a final robot decision action is generated. The method solves the problem that in the robot indoor environment obstacle avoidance task, complex obstacles are difficult to fully perceive, so that obstacle avoidance fails, the robot is helped to utilize semantic information of the environment, interference of redundant pixels is removed, and efficient reinforcement learning training and decision making are carried out. The method has effectiveness and applicability in different scenes.

Description

technical field [0001] The invention belongs to the field of navigation and obstacle avoidance (Navigation and Obstacle Avoidance) in the field of robots. The specific realization result is autonomous navigation and obstacle avoidance of robots, and particularly relates to a method for fully and effectively sensing complex obstacles. Background technique [0002] The robot obstacle avoidance task is that in a more complex scene, the robot can autonomously navigate to the target point without any collision with the obstacle, which has great practical application value. With the rapid development of artificial intelligence technology, tasks related to robot obstacle avoidance, such as sweeping robots, unmanned driving, smart warehouses, smart logistics, etc., have achieved significant performance improvements. [0003] However, there are often some complex obstacles in the indoor obstacle avoidance scene, such as non-convex irregular objects such as tables and chairs, ferrous ...

AI Technical Summary

Problems solved by technology

For laser data, images have rich semantic information, but at the same time have a large amount of redundant information that is not helpful for obstacle avoidance, which makes it difficult for reinforcement learning algorithms to converge and train, and will cause a large gap between virtual and reality, and it is difficult to transfer strategies

And the depth camera has a lot of noise in the indoor environment with sunlight, and it is almost invalid

The traditional method of using the mapping from the depth map to the point cloud to remove the ground interference information is also unable to perceive the lower obstacles on the ground such as clothing and swimming pools.

Therefore, there are many problems in the end-to-end learning method based on RGB-D. It cannot fully perceive the complex indoor environment, let alone navigate and avoid obstacles safely.

Images