Robot neural network type computed torque controller training platform and training method

Abstract

The invention discloses a robot neural network type computed torque controller training platform and training method, belonging to the technical field of robot control. The training platform consists of an upper control unit, an open type motion control driving unit, a high-speed parallel robot virtual reality unit, a data acquisition and communication unit and an artificial neural network-based adaptive control model, wherein motion planning and process monitoring are finished by the upper control unit; the open type motion control driving unit is an actual electrical part which finishes a control command of the upper control unit to convert to electrical control driving; the high-speed parallel robot virtual reality unit is used for realizing construction of a high-speed parallel robot virtual machine and a related three-dimensional scene; the data acquisition and communication unit realizes data communication and can collect a training sample required by the artificial neural network; and the artificial neural network-based adaptive control model consists of a neural network type robot inverse dynamic model approximator and a linear proportional differential feedback controller which work in parallel. The robot neural network type computed torque controller training platform and training method can realize trace tracking control of the high-speed parallel robot and have the characteristics of wide fitness, low system design cost and the like.

Description

Technical field: [0001] The invention belongs to the technical field of robot control, and relates to a training platform and training method for a robot neural network type calculation torque controller, in particular to a high-speed parallel robot neural network type calculation torque controller training platform and training method based on virtual machine and real electricity . Background technique: [0002] The structural characteristics of the parallel robot make it widely used in fields that require high rigidity, high precision or large load without requiring a large working space; in addition, most of its driving devices are placed on or close to the fixed platform , so that the weight of the moving part is light, the dynamic response is good, and high-speed movement can be realized. [0003] In terms of control, at present, most commercial manipulators use linear single-axis control methods such as PD and PID control. Due to the influence of the term, it is impo...

AI Technical Summary

Problems solved by technology

The gap between the simulation-based neural network model and the actual robot model makes it difficult for its controller to be well applied in engineering practice.

On the other hand, if the actual prototype is used to train the neural network, in view of the difference between the initial model and the robot model and the complexity and uncertainty of the training process, it is likely that the device will have an accident during the training process due to the error of the controller model. cost will be unaffordable

Images