Systems and methods of adaptive path planning

Abstract

Systems and methods for providing adaptive path planning techniques using local learning and global planning are described. Adaptive path planning of embodiments provides global guidance and performslocal planning based on local learning, wherein the global guidance provides a planned path through the dynamic environment from the starting location to the selected destination, while the local planning provides dynamic interactions in the environment during arrival at the destination, such as in response to obstacles entering the planned path. Global guidance may combine the initial global pathwith historical information to provide a global path that avoids locations where traffic conflicts often occur. Local planning uses local deep reinforcement learning to guide interaction of an automated vehicle through a global path in a dynamic environment, such as in response to an obstacle entering the global path. A sequential local map may be generated for a deep learning model used by locallearning techniques.

Description

【Technical field】 [0001] The invention relates to adaptive path planning, in particular to an adaptive path planning technology utilizing local learning and global planning. 【Background technique】 [0002] In today's world, autonomous vehicles (AVs) in various forms are becoming more and more popular. For example, AVs in the form of self-guided cars, autonomous delivery vehicles and automated guided vehicles (AGVs) used in warehouses and factories are not uncommon, if not widespread, in industrialized countries. [0003] Path planning algorithms, also known as path finding algorithms, are commonly used in AVs to navigate to desired destinations. Popular path planning methods often implement static search algorithms such as Dijkstra (see "A note on two problems inconnexion with graphs", Numerische Mathematik.1:269-271, Dijkstra, E., et al. The disclosure is incorporated herein by reference) and A* (see "A Formal Basis for the Heuristic Determination of Minimum Cost Paths", ...

AI Technical Summary

Problems solved by technology

However, these learning-based algorithms have not yet provided adequate solutions for path planning in complex, dynamic environments

For example, real-time online path planning methods based on deep and reinforcement learning do not generalize well to very large environments and perform poorly in complex environments

A* and deep heuristics for reinforcement learning also perform poorly in complex environments

Images