Online stably controlled humanoid robot based on bionic reinforcement learning type cerebellum model

Abstract

The invention relates to an online stably controlled humanoid robot based on a bionic reinforcement learning type cerebellum model and belongs to the field of humanoid robots. In order to solve the problem of improving the stabilizing and balancing abilities of the humanoid robot in a walking process, the humanoid robot comprises an apparatus for planning an offline gait of the humanoid robot anda cerebellum model controller, wherein the apparatus output makes the humanoid robot tracks a joint movement locus generated offline to have the walking ability; and the cerebellum model controller, responding to the offline gait, comprises a state encoding module, a cerebellum model, an inferior olive feeding module and a motion mapping module, wherein the state encoding module adjusts an activated state of PF according to state information collected by a sensor of the humanoid robot, the inferior olive feeding module modifies a behavior selection probability and a cerebellum nerve cell storage weight based on evaluation information fed back by an environment, and the motion mapping module adjusts actions of the robot according to output of a functional module. The online stably controlled humanoid robot has an effect of improving the stabilizing and balancing abilities of the humanoid robot in the walking process.

Description

technical field [0001] The invention belongs to the field of humanoid robots, and relates to an online stable control humanoid robot based on a bionic reinforcement learning type cerebellum model. Background technique [0002] A humanoid robot is a robot designed to imitate human appearance and behavior. It constantly replicates the characteristics of human comprehensive capabilities so that it can replace humans to complete repetitive, high-risk, and labor-intensive tasks, so it has broad application prospects. Humanoid robot is a typical multi-joint nonlinear underactuated system, so its gait control is a very challenging problem, and it is also the key to the wide application of humanoid robots. Researchers have proposed a variety of motion gait control methods. At present, the most common method is to decompose the motion task into different sub-modules for planning and control to form an offline gait mode, such as zero point moment (ZMP) theory, inverted pendulum model,...

AI Technical Summary

Problems solved by technology

These cerebellar models are usually only for the purpose of controlling the effect, and although the functional model of the cerebellum is borrowed, the structural properties of the cerebellum are ignored

Images