Brain-inspired Decision-making Method For Rapid Obstacle Avoidance Task Of UAV

UAV is facing more and more complex flight environment,of which the mission types are increasingly expanding.Decision-making of choosing action to avoid fast dynamic obstacles is the basis for UAV to ensure its own safety and complete its tasks in complex environment.However,at present,UAV decision-making is facing challenges such as rapidity and limited airborne resources.It is necessary to explore decision-making methods with faster speed and less resource consumption.To solve this problem,this paper simulates the process of human brain reaching quasi conditioned reflex through training,and explores the brain-inspired decision-making method of UAV for fast obstacle avoidance task.The main research contents and contributions of this paper are as follows:(1)A brain-inspired spiking Actor-Critic(AC)network is proposed based on the study of input-action brain-inspired fast decision-making mechanism of AC network.By analyzing the biological mechanism of AC algorithm and the decision-making mechanism of human brain,and drawing lessons from the early excitation and delayed inhibition from prefrontal cortex to dopaminergic neurons,the current and previous reward estimation can be generated,based on which the temporal difference update method of traditional AC algorithm is improved.Taking advantage of low energy consumption and strong biological interpretation of spiking neural network,a brain-inspired input-action spiking AC network is proposed.(2)A new obstacle avoidance decision-making method for UAV based on spike response model and brain-inspired spiking AC network is proposed.By analyzing the response characteristics of spiking neurons,the spike response model is established.The input-output coding and updating mechanism of brain-inspired spiking AC network is designed,and the decision-making method based on spike response model and spiking AC network is proposed for the first time.The feasibility of the algorithm is verified by the simulation test of flying through window task of quadrotor UAV.(3)A brain-inspired fast decision-making method for UAV facing dynamic obstacle avoidance task is proposed,and the simulation and real flight verification are completed.According to the rapidity requirements brought by dynamic obstacle avoidance decision-making task,the input camera data of spiking AC network are processed by differential coding,and the hidden layer modulation neuron is added to the output timing signal.An asynchronous update mechanism based on A3 C algorithm and a multi-step update mechanism based on λ regression algorithm are studied.A brain-inspired fast decision-making method for UAV facing dynamic obstacle avoidance task is proposed.On XTDrone simulation platform,the decision network is trained and the simulation verification of quadrotor UAV autonomous avoiding flying obstacles is realized.Finally,through the actual flight test of UAV avoiding flying basketball,the accuracy and rapidity of the algorithm are verified.The obstacle avoidance success rate is96% and the average decision-making time is 41.3ms.