Multi-Corridor Lane Change Planning for Autonomous Vehicle Trajectories
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for autonomous vehicle routing fail to accurately determine optimal lane change locations due to insufficient real-time traffic and environmental conditions, leading to potential route failures and inefficiencies.
Innovation Solution
Generate a multi-corridor representation of the local region around the vehicle, incorporating perception data to dynamically determine lane change locations and generate trajectories using static and dynamic costs, allowing for continuous trajectory planning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a single optimal path is searched in the lane graph at the routing stage, then the route planning is simple and fast, but the lane change location determination is inaccurate due to insufficient real-time context
Solution Approach 1:
The patent divides the route planning process into two distinct phases: (1) a routing stage that generates a nominal route using a lane graph, and (2) a trajectory generation stage that refines lane change decisions using a multi-corridor representation. This segmentation allows each phase to focus on its strengths - fast routing vs. accurate lane change positioning - thereby resolving the contradiction between planning efficiency and location accuracy.
Solution Approach 2:
The patent introduces dynamic refinement of lane change locations by using perception data and real-time traffic conditions during the trajectory generation phase. The system dynamically adjusts the nominal route by identifying optimal lane change locations within corridors based on current environmental context, thus improving accuracy without sacrificing the initial routing efficiency.
2Device complexity
If the lane graph is discretized for route representation, then the routing computation is simplified, but feasible lane change windows may be missed or suboptimal locations chosen
Solution Approach 1:
The patent adds a longitudinal dimension to the discrete lane graph by introducing continuous corridors that span multiple lane segments. While the lane graph provides discrete routing options, the corridors extend these options continuously along the route, allowing the system to identify optimal lane change locations within the corridor length rather than being constrained to discrete graph nodes. This dimensional extension maintains computational simplicity while improving lane change reliability.
3Productivity
If lane change decisions are made without real-time perception data, then the routing process is faster, but the decisions may be inaccurate or infeasible
Solution Approach 1:
The patent performs preliminary routing to generate a nominal route and identify potential corridors before using perception data for precise lane change location selection. This preliminary action establishes a foundation that guides subsequent perception-based refinement, allowing the system to leverage both fast routing algorithms and accurate real-time sensing without compromising either speed or precision.
Solution Approach 2:
The system uses perception data as feedback to refine the nominal route generated during routing. The trajectory generation phase continuously monitors real-time conditions (traffic, obstacles, road geometry) and adjusts lane change locations within corridors based on this feedback, ensuring decisions are both fast and accurate by combining pre-computed routing with real-time adaptation.
4Reliability
If the AV waits to be re-routed when unable to make a planned lane change, then routing safety is maintained, but unexpected stops occur reducing productivity
Solution Approach 1:
The patent performs preliminary identification of multiple feasible lane change locations within corridors during the trajectory generation phase, before the vehicle reaches the routing decision point. By pre-identifying alternative locations within the same corridor, the system ensures that if one lane change opportunity is missed or blocked, the vehicle can immediately execute a lane change at an alternative location within the corridor rather than waiting for a complete re-route, thereby maintaining safety while improving travel efficiency.
Data Source
AI summary
Methods and systems for maneuvering an autonomous vehicle are disclosed. The methods include generating a multi-corridor representation corresponding to a local region around the autonomous vehicle while travelling on a route, and using the multi-corridor representation and perception data corresponding to the local region to generate a trajectory for the autonomous vehicle to traverse the local region. The multi-corridor representation includes a plurality of adjacent corridors that each include one or more lane segments of a road network. A location of executing a lane change along the route is determined dynamically during a trajectory generation phase based on the perception data.