Research on machine vision decision-making system for self-driving logistics vehicles

Aiming at the integrated problem of perception and decision-making of self-driving logistics vehicles in complex traffic environments, this study starts from four aspects: environment perception and cognition, scene understanding and analysis, intelligent decision-making and obstacle avoidance, and optimized path planning. Deep convolutional neural network models are utilized to extract, analyze and understand sensory data, and multi-sensor data fusion technology is used to integrate the information. Reinforcement learning methods are applied for the system to automatically learn decision-making strategies, while path planning algorithms are used to find the optimal driving path. In the simulation experiments of the machine vision decision-making system using Gazebo, the recognition accuracy reaches 98.2%. The results demonstrate the feasibility of the technical solution for an automated driving machine vision decision-making system.