Autonomous Vehicle Path Planning Using Human Driving Priors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing approaches for deriving planned paths for autonomous vehicles (AVs) often lead to unnaturalistic driving behavior, as they primarily rely on geometric lane markings and boundaries, neglecting human driving behavior, which can degrade the experience of passengers.
Innovation Solution
The technology involves a two-phase process: a learning phase where actual human-driving paths are recorded and incorporated into map data, and a runtime phase where these path priors are used to derive a more naturalistic planned path for AVs by filtering, sampling, and fitting curves to produce a smooth trajectory.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If geometric lane markings and boundaries are used as the primary basis for deriving planned paths, then the path derivation process is simple and straightforward, but the resulting driving behavior becomes unnaturalistic and degrades passenger experience
Solution Approach 1:
The system performs preliminary action by collecting and storing path prior data from human-driven vehicles during normal operations. This pre-collected data is stored in map data structures, enabling the autonomous vehicle to reference realistic human driving patterns without requiring complex real-time data collection or processing, thus maintaining operational simplicity while improving behavior naturalism
Solution Approach 2:
The system uses copying by replicating actual human driving paths (path priors) from human-driven vehicles and storing them as reference data. These copied paths are then used during autonomous operation to generate planned paths that mirror natural human behavior, replacing the need for complex geometric calculations while preserving driving naturalism
2Reliability
If path priors from human driving are collected and processed through filtering, sampling, and curve fitting, then driving behavior becomes more naturalistic, but the system complexity increases
Solution Approach 1:
The system applies segmentation by dividing the path prior processing into distinct modular stages: filtering (removing irrelevant path priors based on contextual conditions), sampling (selecting representative path points), and curve fitting (generating smooth planned paths). Each stage is independently implemented and can be processed separately, reducing overall system complexity while achieving naturalistic driving behavior
Solution Approach 2:
The system uses an intermediary approach by introducing map data as a intermediate storage layer between path prior collection and path derivation. The map data stores pre-processed path prior information organized by geographic location, serving as a mediator that simplifies real-time path planning by providing ready-to-use reference paths without requiring complex real-time processing
3Measurement precision
If contextual filtering is applied to path priors based on time, weather, and traffic conditions, then the accuracy of planned paths improves, but the processing time and computational load increase
Solution Approach 1:
The system performs preliminary action by pre-filtering and organizing path priors during data collection phases, storing only relevant path information in map data structures organized by contextual conditions (time of day, weather, traffic). This pre-processing eliminates the need for extensive real-time filtering during path planning, maintaining high accuracy while reducing computational time
Solution Approach 2:
The system applies local quality by selectively applying different filtering criteria and path prior selection strategies based on specific contextual conditions and geographic locations. Rather than uniformly processing all path priors with the same complexity, the system adapts the processing level to local requirements, improving accuracy where needed while minimizing processing time in less critical scenarios
Data Source
AI summary
Disclosed herein is technology for deriving a planned path for a vehicle that is operating in a geographic area. In an example embodiment, a computing system associated with the vehicle may function to (i) identify a set of path priors to use as a basis for deriving the planned path for the vehicle in the geographic area, (ii) sample path priors in the identified set of path priors and thereby producing a set of points that is representative of the identified set of path priors, (iii) fit a curve to the set of points produced by the sampling, and (iv) derive the planned path for the vehicle based on the fitted curve.