Real-Time, Secure, and Computationally Efficient Navigation: A Prediction-Planning Framework Using Occupancy Grid Map

Real time navigation is an important part of the intelligent agent's ability to drive safely in complex dynamic environments, which consists of prediction and motion planning. However, most existing methods require high-quality sensor system and gigantic computation consumption, which oppose the real-time requirements. Moreover, these methods lack a unified representation framework for prediction and planning, arising loss of accuracy during representation conversions. To address these issues, we propose a prediction-planning framework based on an occupancy grid map. Prediction module combines perception information with motion information to handle dynamic and static obstacles and generate a sequence of occupancy grid maps. Planning module samples, selects, and optimizes candidate trajectories. The experimental results in multiple scenarios demonstrate that the proposed framework can enhance real-time performance and security, while reducing computational requirements compared to previous methods.