Research On Obstacle Avoidance Navigation Algorithm Of Intelligent Distribution Robot

In recent years,with the increase in delivery tasks brought about by online shopping and the need for isolation due to the spread of the epidemic in 2020,society has also paid more and more attention to unmanned delivery tools.With the continuous development of artificial intelligence technology,the application of intelligent delivery robots has gradually matured and is widely used in daily life scenarios such as express delivery and takeaway.In addition,during the epidemic,the demand for delivery robots in medical scenarios has become more and more urgent.In addition to reducing the amount of labor for medical staff,it can also avoid cross-infection between staff to a certain extent.This article mainly focuses on the difficulties of autonomous delivery robots in obstacle avoidance navigation technology.In order to deal with practical problems such as the dense flow of indoor obstacles and complex types,the SAC(soft actor-critic)in deep reinforcement learning(DRL)is used.Based on the algorithm,the model of the obstacle avoidance navigation problem of the distribution robot is studied.The main contents are as follows:(1)This article proposes a parallel SAC(PSAC)training algorithm to train the robot's obstacle avoidance navigation ability in response to the problems of dense and complex obstacles in actual application scenarios.This speeds up the training while speeding up the training.It also improves the generalization performance of the model to meet the flexibility and stability needs of the robot.In the model testing phase,the performance difference between the SAC algorithm and the traditional artificial potential field(APF)algorithm was compared.Finally,the trained model is transferred to a real robot with RGBD camera as the observation sensor.In the experiment,the robot can flexibly avoid dynamic pedestrians.The result proves that the algorithm proposed in this paper has strong practicability.(2)Aiming at the problem that a single sensor cannot obtain complete observation information due to the complex types of obstacles,this paper proposes a data fusion method based on deep learning for RGBD cameras and ultrasonic sensors.It can be easily modified by modifying training data or network structure.The data fusion algorithm is incorporated into the SAC obstacle avoidance algorithm.Experiments show that after data fusion,the robot can have good stability in light-transmitting or opaque obstacle scenes.(3)For long-distance obstacle avoidance navigation tasks,this paper proposes an RRT-SAC obstacle avoidance navigation algorithm combined with RRT(Rapidly exploring Random Tree).First use the RRT algorithm for path planning,divide the long path into several short paths,and then call the SAC algorithm to handle short-distance obstacle avoidance navigation tasks.Experiments show that the RRT-SAC method can effectively solve the obstacle avoidance navigation problem of long paths.The research in this article on the obstacle avoidance navigation technology of distribution robots can provide references for further research in this field.