Vision robot motion control method based on reinforcement learning

Abstract

The invention discloses a vision robot motion control method based on reinforcement learning, and belongs to the technical field of robot control. The vision robot motion control method based on reinforcement learning comprises main imaging data collecting, branch sub-information collecting, range space model establishing, motion track strategy formulating, branch information periodic transmissionand real-time motion path correction; supplementing is conducted on an established range space model according to the path changing information collected by a branch sub probe in the fifth step, themotion track information is corrected in real time, supplementing can be conducted on the robot vision field range through the reinforcement learning algorithm and an outer machine vision probe, the motion track is adjusted in time, the motion track deviation value is reduced, and the robot motion track running correct rate is increased; and meanwhile, self-checking is conducted on robot exteriorchanges, and robot motion control influences from corrosion caused by the external environment are reduced. Vision field sharing and touch vision supplementing are achieved.

Description

technical field [0001] The present invention relates to the technical field of robot control, and more specifically, relates to a visual robot motion control method based on reinforcement learning. Background technique [0002] Vision robot refers to not only taking visual information as input, but also processing the information, and then extracting useful information to provide to the robot. The main working principle is: three-dimensional objects in the objective world are transformed into two-dimensional objects The plane image of the object is then processed to output the image of the object. Usually, robots need two types of information to judge the position and shape of objects, namely distance information and light and shade information. Of course, as object visual information, there is also color information, but it is not as important as the first two types of information for object position and shape recognition. The robot vision system is very dependent on ligh...

AI Technical Summary

Problems solved by technology

[0005] Although the above technical solution solves the problem of integrating machine vision and robot controllers into a unified controller, in the process of robot motion control, the robot needs to first locate its own position and attitude information, and then understand the target position information through machine vision, and then through The internal system calculates the motion trajectory to calculate the correct movement method, including the rotation or swing range of the robot's own multiple motion axis joints. However, some robots work in special environments, such as seabed or high-corrosion environments. Due to the existing robot limbs Most of them are made of metal. In special environments, the outer layer of the metal is susceptible to corrosion reactions, resulting in changes in the posture of the machine itself, and even structural deformation. In this case, the correct motion trajectory of the robot deviates from the normal calculated motion trajectory. , the displacement accuracy decreases

Images