Trajectory Optimization for Autonomous Mobility Platforms through Neural-Based Perception Models

This paper introduces an enhanced Deep Reinforcement Learning-based Path Planning (DRL-PP) algorithm for autonomous driving systems. A* and Dijkstra's algorithm are two common ways to plan a path. They work well in static environments, but they don't work as well in real-world situations that are dynamic and uncertain. To get around these problems, the suggested method uses deep reinforcement learning to make autonomous vehicles' decision-making more flexible and smart. The DRL-PP algorithm employs neural networks to derive optimal actions from environmental states and to formulate efficient pathways from the origin to the destination. One of the most important things this work does is improve the reward function by adding things like avoiding collisions, making the path smoother, making travel more efficient, and getting closer to the goal. This dynamic reward system helps you choose the right action, keeps training stable, and stops overfitting. Results from experiments.