Gait training method and device of quadruped robot based on deep reinforcement learning, electronic equipment and medium

Abstract

The invention discloses a gait training method and device of a quadruped robot based on deep reinforcement learning, electronic equipment and a medium. According to the application, an environment, the quadruped robot model and a sensor model can be built in the simulator; a measurement signal is acquired through a multi-source information fusion technology and a sensor, and the state of the robotis estimated; a foot end track planning and initial gait control strategy of the quadruped robot is generated by utilizing the environment module, the robot model and the measurement signal; and strategy improvement is performed on the initial gait control strategy of the quadruped robot based on a deep reinforcement learning algorithm until a strategy improvement ending condition is triggered. By applying the technical scheme of the invention, the problems that a large amount of manual parameter adjustment is needed and the labor cost is relatively high due to a model control algorithm in related technologies can be avoided. Parameters obtained through training are often poor in robustness, can only be suitable for a certain fixed scene, and are difficult to meet the requirements of various application tasks of an actual quadruped robot.

Description

technical field [0001] This application relates to computer technology, especially a gait training method, device, electronic equipment and medium for a quadruped robot based on deep reinforcement learning. Background technique [0002] In the field of gait control of quadruped robots, methods such as forward and inverse kinematics and robot dynamics are usually combined with advanced control theories such as model predictive control to realize the gait control of quadruped robots. The entire control system is usually divided into foot trajectory planning Module and motion control module, the planning part realizes the calculation of the trajectory of the foot end, and the control part realizes the control of each joint actuator according to the trajectory. However, the method based on such traditional control theory requires a lot of manual parameter adjustment, and the noise and delay of sensor information from a single source are relatively large, which is likely to cause...

AI Technical Summary

Problems solved by technology

However, the method based on such traditional control theory requires a lot of manual parameter adjustment, and the noise and delay of sensor information from a single source are relatively large, which is likely to cause problems in the control system, and the robustness of the model is difficult to meet the needs of complex environments.

In addition, when the quadruped robot encounters uneven and unstructured terrain, the traditional step control strategy is prone to failure. The fundamental reason is that the model assumptions of such methods are strong, and it is difficult to meet the complex, highly dynamic, and multi-scenario requirements of the actual quadruped robot. Usage requirements

[0003] At present, the research on the gait control of quadruped robots in related technologies is still relatively simple, and most of them focus on the simulator based on the physics engine and the algorithm level of reinforcement learning, lacking a complete starting from the simulator and transferring to the physical robot feasible method

Images