AV Planning Reference Priors From Human Driving Nominal Paths
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing autonomous vehicle (AV) systems face challenges in mimicking human driving behavior, particularly in areas of high curvature or complex shapes, leading to non-human-like driving behaviors that can compromise safety, and the availability and scalability of manual driving data from AV technicians (AVTs) are limited.
Innovation Solution
The use of manual driving data from trained AVTs to generate updated centerlines (MDD nominals) and algorithmic nominals, such as smoothing splines or Iterative Linear Quadratic Regulator (ILQR), to create geometrically optimal reference paths that emulate human-like driving, supplemented by AVT-generated data where available, and applied where human-like driving is necessary.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If geometric centerlines are used for path planning, then the path planning is simple and efficient, but the driving behavior becomes non-human-like and unsafe in high curvature areas
Solution Approach 1:
The system pre-generates nominal reference paths through manual driving data collection and algorithmic processing before actual autonomous driving operations. These pre-computed paths serve as priors that guide the planning stack, eliminating the need for real-time complex calculations while ensuring human-like driving behavior in high curvature and complex shape areas.
Solution Approach 2:
The patent introduces nominal reference paths as an intermediary layer between the map data and the planning stack. These nominal paths act as a mediator that translates geometric centerlines into human-like driving trajectories, resolving the contradiction between simplicity and safety by providing pre-processed guidance paths.
2Reliability
If manual driving data is collected to generate updated centerlines, then more accurate human-like paths are achieved, but the system complexity and data processing requirements increase
Solution Approach 1:
Manual driving data is collected and processed in advance to generate nominal reference paths before deployment. This preliminary action allows complex data processing to be performed offline, reducing real-time system complexity while maintaining high path accuracy through thorough pre-computation.
Solution Approach 2:
The system uses algorithmic nominal generation that automatically processes manual driving data without requiring continuous human intervention. Once the nominal path generation system is established, it self-services by automatically computing reference paths from collected data, reducing operational complexity while maintaining accuracy.
3Reliability
If manual driving data is used for all map areas, then comprehensive coverage is achieved, but the data availability and scalability are limited
Solution Approach 1:
The system performs preliminary algorithmic nominal generation for areas where manual data is unavailable or insufficient. By pre-computing nominal paths using algorithms for regions lacking manual data, the system achieves comprehensive map coverage while adapting to varying data availability across different geographic areas.
Solution Approach 2:
The patent creates a universal nominal path generation system that can handle both manually-data-rich areas and data-scarce areas through different methods. The same infrastructure supports multiple data sources and generation approaches, making the system adaptable to varying data availability across different map regions.
Data Source
AI summary
A method for providing a reference prior for a portion of a route to a planning stack of a vehicle. The method includes accessing manual driving data associated with the route portion, the manual driving data stored in a database; averaging the accessed manual driving data to determine a nominal path for the route portion; and providing the nominal path to the planning stack of the vehicle as the reference prior for the route portion, wherein the planning stack consumes the reference prior for planning maneuvers of the vehicle in traversing the route portion.