Variable-Dimensional Trajectory Planning for Longer AV Horizons
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing trajectory planning systems for autonomous vehicles face computational challenges that restrict their planning horizon, leading to inadequate long-term planning and potential planning or control failures, especially in maneuvers like lane changes.
Innovation Solution
A multi-dimensional planning system that dynamically adjusts dimensionality over time, reducing the number of active dimensions at each stage to minimize computational effort while extending the planning horizon, allowing for safer and more proactive vehicle maneuvers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Duration of action of moving object
If the planning horizon is extended to enable longer-term trajectory planning, then the vehicle can anticipate hazards earlier and perform maneuvers more proactively, but the computational load increases significantly
Solution Approach 1:
The planning process is divided into multiple stages, where each stage operates with a specific dimensionality. The first stage uses high dimensionality for short-term planning, while subsequent stages use reduced dimensionality for long-term planning. This segmentation allows the system to extend the overall planning horizon without proportionally increasing computational load at each step.
Solution Approach 2:
The system dynamically adjusts the dimensionality of the planning process based on the stage and time horizon. For near-future planning, all dimensions are active to capture detailed vehicle dynamics. For far-future planning, only essential dimensions are maintained while others are reduced or fixed, allowing computational resources to scale with the planning horizon rather than remaining constant.
2Measurement precision
If high dimensionality is maintained throughout the planning process to capture all vehicle dynamics, then planning accuracy is improved, but computational complexity becomes prohibitive for long-term planning
Solution Approach 1:
Different dimensionalities are applied to different time horizons and planning stages. For immediate future planning where precision is critical, full dimensionality is maintained. For distant future planning where rough trajectory guidance suffices, reduced dimensionality is used. This local differentiation of quality allows accurate local planning without globally high computational complexity.
Solution Approach 2:
The system applies full dimensional planning only when necessary (short-term, critical maneuvers) rather than continuously. For long-term planning, a partial dimensional approach is used that captures essential dynamics while omitting less critical details. This selective application of full action where needed resolves the contradiction between comprehensive accuracy and overall computational feasibility.
Data Source
AI summary
A multi-dimensional trajectory planning system is disclosed that includes planning and actuator modules. The planning module executes the planning application to: determine a first dimensionality including first dimensions for a first stage, where each of the first dimensions are active, and where the first dimensions include two or more dimensions; determine a second dimensionality including second dimensions for a second stage, where the second dimensions include the first dimensions or a subset of the first dimensions, and where the second stage has a lower level of dimensionality than the first stage; based on map data and sensor data, estimates first possible future states of the first dimensions for the first stage, and estimates second possible future states of the second dimensions for the second stage based on the first possible future states; and selects a trajectory plan based on the second possible future states.