Autonomous Vehicle Routing Policy via Lane Graph Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing route planning mechanisms for autonomous vehicles, which represent road networks as directed graphs, become impractical due to the large number of edges, making classical shortest path algorithms like Dijkstra inefficient for determining optimal routing policies.
Innovation Solution
The generation of a lane graph with vertices representing road portions and associated costs for stay-in-lane, lane change, and forced lane change actions, allowing for the application of a shortest path first algorithm to determine an optimal routing policy, considering the probability of lane change success and costs to discourage undesirable behaviors like frequent lane changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If road network is represented as directed graph with detailed edges for each road segment, then routing accuracy is improved, but computational complexity increases making classical shortest path algorithms impractical
Solution Approach 1:
The patent segments the continuous road network into discrete lane segments with specific properties. Each lane segment is represented as a node in a simplified graph, with attributes like lane type, speed limit, and connectivity. This segmentation reduces the complexity from millions of edges to a manageable number of lane segment nodes while preserving essential routing information for autonomous vehicle navigation.
Solution Approach 2:
The patent extracts only the essential routing features from the complete road network data. Instead of representing every possible vehicle trajectory and lane change as separate edges, it extracts key lane segments and their connectivity relationships. This extraction creates a simplified graph structure that captures the necessary routing information without the computational burden of the full detailed network.
2Reliability
If classical shortest path algorithms like Dijkstra are applied to complete road network graph, then optimal route is found, but computation time becomes unacceptable due to millions of edges
Solution Approach 1:
The patent segments the road network into hierarchical levels (road segments, lane segments, maneuver segments) that can be processed efficiently. By organizing the graph structure at appropriate levels of abstraction, it enables the application of shortest path algorithms on a reduced graph size while still finding optimal routes for autonomous vehicle navigation.
Solution Approach 2:
The patent changes the parameters of the graph representation by using lane segments as nodes instead of individual road segments or intersections. It introduces new attributes like lane connectivity scores and maneuver probabilities that allow for more efficient pathfinding. These parameter changes enable the use of modified shortest path algorithms that run faster on the transformed graph structure.
Data Source
AI summary
Methods and systems for determining an optimal routing policy for an autonomous vehicle are described. A set of states that represent portions of lanes are determined. A set of costs for a set of actions is associated with each state. The set of actions includes a stay in lane action, a lane change action, and a forced lane change action. A lane change action has a non-deterministic outcome with a probability based on the length of the state. A cost for a stay in lane action is greater or equal than a length of the state and a cost of a forced lane change action is the reciprocal of a success rate of lane changes. An optimal routing policy is determined for the autonomous vehicle according to a shortest path first (SPF) algorithm run on the set of states based on the set of costs and the set of actions.