Modeling And Control Of Underwater Bionic Octopus Biped Robot Based On Deep Reinforcement Learning

At present,land resources have been extensively developed,The undeveloped ocean is regarded as the target of resource development,and underwater robots are used to explore and inspect underwater resources.Most of the traditional underwater robots are rigid,which have the disadvantages of insufficient mobility and poor stability.However,The marine cephalopod octopus has the characteristics of high swimming efficiency,better environmental adaptability and locomotion ability,which provide a source of bionic inspiration for underwater robots.However,the complex working environment and dynamic characteristics of the underwater bionic robot bring challenges to its motion control.Therefore,this paper focused on analyzing the kinematics,dynamics modeling and control system of the underwater bionic robot.In this paper,the octopus was taken as the research object of the underwater robot bionic research,and the modeling and control of the cable-driven bionic octopus arm were deeply analyzed and developed.Based on the underwater spring loaded inverted pendulum model,gait modeling analysis of the underwater bionic octopus biped robot was completed,and the parameters of the legged locomotion model by RNA genetic algorithm were optimized.The specific research work is as follows:(1)The structure design and modeling of bionic octopus arm were analyzed.In this paper,the cable-driven bionic octopus arm was used as the motion mechanism of the bionic octopus bipedal robot prototype,and the development of the bionic octopus arm were completed.The multi-layer perceptron neural network method driven by experimental data was used to model the bionic octopus arm,and the relationship between the pulling and the shape of the driving line of the bionic octopus arm was designed.(2)The control strategy of the end position of the bionic octopus arm was designed.The control strategy of bionic octopus arm based on deep q-network was designed.The design of state space,the definition of action space,the design of state transition strategy and reward function of the bionic octopus arm were analyzed respectively.The variation of the effectiveness and robustness of training of the end position control of the bionic octopus arm was completed by DQN algorithm in the environment of simulation and experiment.(3)The parameters of legged locomotion model of underwater bionic octopus biped robot were optimized.According to the characteristics of the octopus arm rolling contact with the ground,the biped motion process of the underwater bionic octopus robot was analyzed.Based on the U-SLIP model,the gait modeling of underwater bionic octopus biped robot was completed.RNA genetic algorithm was used to optimize the parameters of legged locomotion model and improve the problems of short walking distance and motion instability of underwater bionic octopus biped robot.The effectiveness of the algorithm for legged locomotion model optimization was verified by simulation.