Robot feedforward force moment compensation method

Abstract

The invention relates to a robot feedforward force moment compensation method. The robot feedforward force moment compensation method comprises the following steps that S1, a motion-value neural network is built; S2, a training track is generated; S3, according to the state at the current moment, corresponding motions are selected, integral of the feedforward force moment increments of the selected motions is carried out, the feedforward force moment increments are output to a motor current ring feedforward passage, and immediate return at the current moment and the next-moment state are obtained; S4, the state of the current moment, the selected feedforward force moment increments, the immediate return and the next-moment state serve as neural network training samples, and the training samples are stored into a queue after being subject to normalization; and S5, parts of training samples are selected from the queue at random, and a stochastic gradient descent method is used for training a motion-value neural network until the maximum training time number is obtained or joint tracking errors are smaller than an error threshold value. According to the method, a complex kinetic modeldoes not need to be built, real-time compensation of industrial robot joint force moment can be achieved, and high-accuracy control is achieved.

Description

technical field [0001] The invention relates to the technical field of robot control, in particular to a method for robot feedforward torque compensation. Background technique [0002] With its flexibility, versatility, high precision and low cost, robots have become one of the most widely used equipment in the field of construction machinery manufacturing. With the continuous expansion of robot application fields and the rapid development of modern industry, high speed and high precision have become the main development trend of robots, and robot torque feedforward compensation is the key to improving its motion accuracy. Therefore, accurate torque feed-forward compensation is of great significance to realize high-speed and high-precision control of robots. [0003] The feed-forward torque compensation of robots generally adopts the dynamic model to calculate the feed-forward torque. This method can achieve a good compensation effect in theory, but in practical application...

AI Technical Summary

Problems solved by technology

[0003] The feed-forward torque compensation of robots generally uses a dynamic model to calculate the feed-forward torque. This method can achieve a good compensation effect in theory, but in practical applications, it is often difficult to obtain ideal results.

The common problems are: (1) The algorithm relies too much on the accuracy of the dynamic model; (2) There are too many dynamic model parameters, which are difficult to identify; (3) During the long-term operation of the robot, the dynamic The accuracy of the model is reduced, or even invalid

Images