Autonomous Vehicle Path Planning for Narrow Multi-Obstacle Lanes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Self-driving vehicles struggle to plan paths in narrow spaces and effectively avoid multiple obstacles due to methods like obstacle inflation and rule-based determinations failing to provide optimal and smooth paths.
Innovation Solution
A path planning method that projects lane lines, center-line, and obstacle contours into a target coordinate system, performs sampling on both sides of obstacles to obtain points, calculates costs based on distances and smoothness, and determines a reference path to ensure proximity to the lane center-line while avoiding obstacles.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If traditional path planning algorithms (A*, Dijkstra) are used, then path finding capability is provided, but computational complexity increases and real-time response is difficult to achieve
Solution Approach 1:
The patent segments the path planning problem into two distinct phases: a global coarse path planned by A* algorithm, and local fine-path adjustments handled by reinforcement learning. This segmentation allows each component to focus on specific aspects, reducing overall computational complexity while maintaining planning speed.
Solution Approach 2:
The patent introduces an intermediary reinforcement learning model that acts as a bridge between the global path planner and the robot executor. This intermediary processes the coarse path and generates optimized local paths in real-time, resolving the contradiction between global optimality and local responsiveness.
2Manufacturing precision
If reinforcement learning is used for path optimization, then local path optimization is achieved, but training time and computational resources increase
Solution Approach 1:
The reinforcement learning model is trained in advance during an offline phase to learn optimal local path adjustment strategies. Once trained, the model can perform real-time path optimization without requiring additional training time during actual robot operation, thus achieving high precision without time loss during deployment.
Solution Approach 2:
The patent applies reinforcement learning selectively only for local path adjustments rather than entire path planning, using partial action to optimize only the necessary portions of the path where dynamic obstacles or narrow passages exist, thereby reducing overall computational burden while maintaining precision.
3Adaptability or versatility
If global path planning algorithms are used, then complete path coverage is achieved, but real-time adaptation to dynamic obstacles is poor
Solution Approach 1:
The patent transforms the static global path into a dynamic system by introducing a reinforcement learning model that continuously adapts the path in real-time based on current environmental conditions and dynamic obstacles. This dynamic adjustment mechanism maintains path execution reliability while improving adaptability to changing conditions.
Solution Approach 2:
The system implements feedback mechanisms where the reinforcement learning model continuously receives information about dynamic obstacles and path execution status, then adjusts the local path accordingly. This closed-loop feedback ensures reliable path execution while adapting to real-time changes in the environment.
Data Source
Figure 1~2
Figure 3
Figure 4~5
AI summary
A path planning method and apparatus, an electronic device and a storage medium, relating to the field of intelligent transportation. The method includes: projecting (S301) information on lane lines, information on a lane center-line and information on an obstacle contour into a target coordinate system to obtain projection information; determining (S302) a start point and an end point of path planning in the target coordinate system, and if it is determined based on the projection information that there is an obstacle between the start point and the end point, then performing sampling on both sides of the obstacle contour along a travelling direction of a vehicle to obtain sampling points; determining (S303) a plurality of paths based on the start point, the end point and the sampling points; performing (S304) a cost calculation on the plurality of paths, and determining a reference path in the plurality of paths based on results of the cost calculation; and determining (S305) a target planned path based on the reference path. The path planning method simplifies a sampling space, realizing that a self-driving vehicle can effectively and reasonably avoid multiple obstacles in a narrow travelling space with ensured optimality and smoothness of the path.