Robot obstacle avoidance device, obstacle avoidance control method and robot

Abstract

The invention relates to a robot obstacle avoidance device, an obstacle avoidance control method and a robot. On the one hand, time series features and spatial features are extracted, and target position data and robot speed data are integrated, which can improve the feature expression ability of deep reinforcement learning and improve the value of preset actions. Predict timeliness and accuracy; on the other hand, fully consider the scene data of the robot from t-T1 to t, and effectively use the current local environment information, historical information and spatial information to accurately evaluate the value of the robot's preset actions, which can ensure the robot The reliability of subsequent obstacle avoidance is significantly improved compared with the existing technology, and the accuracy and timeliness of its obstacle avoidance effect are significantly improved, and the original lidar ranging data can be directly used, which has fast response ability and is convenient for network operation. Deployment and application.

Description

technical field [0001] The invention relates to the field of robot control, in particular to an obstacle avoidance control of a robot. Background technique [0002] With the development of robotics, robots have been widely used in various scenarios, such as courier robots, sweeping robots, and security robots. It uses sensors such as lidar and cameras to perceive the surrounding environment, thereby achieving collision avoidance and autonomous navigation. However, in a dynamic environment, the interactive movement behaviors from vehicles, pedestrians, and other robots bring great challenges to the robot's obstacle avoidance. [0003] In recent years, obstacle avoidance methods based on deep reinforcement learning have excellent performance in dynamic environments, and are currently a hot research topic. The published patent CN108255182B has developed a pedestrian perception obstacle avoidance method for service robots based on deep reinforcement learning, which can integra...

AI Technical Summary

Problems solved by technology

[0004] These methods have achieved good results, but when constructing deep reinforcement learning neural networks, they usually only consider the current local environmental information, lack of effective use of historical information and spatial information, and the obstacle avoidance effect in dynamic and complex environments is not good. to be improved

Images