Autonomous Parking Path Planning With Neural Node Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing pathfinding algorithms for autonomous vehicle parking, such as Dijkstra and heuristic algorithms, suffer from exponential computational time increases as the number of nodes increases, making them inefficient in complex parking scenarios with obstacles.
Innovation Solution
Integrate a neural network trained via reinforcement learning to predict optimal movement paths, reducing the number of nodes analyzed by prioritizing the least costly paths and using traditional algorithms iteratively to handle obstacles.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional pathfinding algorithms (Dijkstra, A*) are used to determine optimal vehicle paths, then path optimality is achieved, but computational time increases exponentially as the number of nodes increases
Solution Approach 1:
The patent replaces traditional mechanical pathfinding algorithms (Dijkstra, A*) with a neural network-based system. The neural network is trained via reinforcement learning to predict optimal paths directly, substituting the computational mechanics of traditional algorithms with a learned predictive model that generalizes across different parking scenarios, thereby reducing computational time while maintaining path optimality.
Solution Approach 2:
The neural network is trained in advance through reinforcement learning on simulated parking environments with various obstacles and configurations. This preliminary training allows the network to encode optimal pathfinding strategies beforehand, so during actual vehicle operation, the pre-trained network can quickly predict paths without performing exhaustive searches, thus resolving the time-complexity contradiction.
2Reliability
If exhaustive search methods are used to ensure kinematic feasibility and obstacle avoidance, then path reliability is improved, but computational complexity increases
Solution Approach 1:
The patent substitutes traditional exhaustive search methods with a neural network that has been trained to implicitly understand kinematic constraints and obstacle avoidance. The network learns feasible paths through reinforcement learning in simulated environments, replacing complex computational searches with a trained predictive model that generalizes to new scenarios while maintaining reliability.
Solution Approach 2:
The patent changes the approach from discrete algorithmic parameter adjustments to continuous learned parameters. Instead of using fixed algorithms with discrete decision points, the neural network learns continuous path representations and parameters through reinforcement learning, allowing it to adapt to varying obstacle configurations and kinematic constraints without increasing computational complexity.
Data Source
AI summary
A method of determining a movement path for a vehicle includes determining a set of adjacent nodes with respect to a current node. The method includes determining, via a neural network, a subsequent path node and determining whether a path based on the subsequent path node connects a source to a goal node. The method includes, in response to a determination that the path based on the subsequent path node does not connect the source node to the goal node, determining an alternative subsequent path node, and determining, via the neural network, a second path based on the alternative subsequent path node. The method includes, in response to a determination that the path based on the subsequent path node connects the source to the goal node, selecting the path based on the subsequent path node as the movement path, and executing a set of actions associated with the movement path.