Path Planning Method for Manipulator Based on Velocity Smooth Deterministic Policy Gradient

Abstract

The invention discloses a path planning method of a manipulator based on a smooth deterministic strategy gradient of speed, and constructs a manipulator simulation environment with job task feedback in the training stage; introduces the previous step manipulator action vector into the deterministic strategy gradient network input, Construct a reinforcement learning network framework based on velocity smooth deterministic policy gradient; initialize network training parameters and manipulator simulation environment; obtain samples based on velocity smooth deterministic policy gradient network and simulation environment, and build a training sample library. If the number of training samples reaches the maximum number of samples number, the training samples are drawn from the training sample library according to the number of single training samples, and the training speed smoothes the deterministic policy gradient network; otherwise, proceed to the next step or the next simulation. On the basis of the deterministic strategy gradient network, the present invention adds the velocity vector of the previous step as the network input, effectively reduces the joint acceleration, and reduces the vibration of the mechanical arm.

Description

technical field [0001] The invention relates to a path planning technology of a manipulator, in particular to a path planning method of a manipulator based on velocity smoothing deterministic policy gradient. Background technique [0002] With the continuous development of robot technology, robotic arms have gradually replaced humans in special industries such as construction, palletizing, medical treatment, and live work. The traditional method of teaching and controlling the manipulator cannot meet the requirements of complex application scenarios and application tasks because it relies on manually given paths. Therefore, the autonomous path planning technology of the manipulator is a key technology for the development of intelligent robots. [0003] At present, the autonomous path planning method of the manipulator based on sampling path planning and multi-objective optimization can adapt to different operating environments, but it cannot be planned online, and the plann...

AI Technical Summary

Problems solved by technology

However, since there is no smoothing process, the output motion vector will be quite different at the front and rear moments, resulting in excessive acceleration, which will cause the mechanical arm to vibrate and cause damage to the mechanical arm

Images