Lane Graph Geometry Search for Unmapped Self-Driving Navigation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Self-driving vehicles face challenges in navigating areas without pre-mapped roads and dealing with inaccuracies in existing maps due to new obstacles or changes, such as construction, which can lead to navigation errors.

Innovation Solution

The techniques generate lane graph topologies and geometries deterministically, using a search method that scores and ranks possible lane graphs to select the most likely one, allowing for real-time computation and integration of external evidence like road signs, without relying on machine learning, ensuring fully autonomous navigation even in unmapped areas.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If pre-generated maps are used for navigation, then navigation efficiency is improved, but navigation accuracy deteriorates in unmapped or changed areas

Engineering Contradiction:
Improvenavigation efficiencyVSAvoidnavigation accuracy
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The system performs preliminary actions by pre-generating maps for efficient navigation in known areas, while simultaneously preparing real-time lane graph generation capabilities to handle unmapped or changed areas, thus resolving the contradiction between navigation efficiency and accuracy

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system uses the self-driving vehicle's own sensor data to generate and update lane graphs in real-time, making the system self-sufficient for handling unmapped areas without relying on external map updates, thereby maintaining navigation accuracy in dynamic environments

Inventive Principle:
Principle #25Self-service

2Adaptability or versatility

If machine learning techniques such as deep neural networks are used, then adaptability to unknown regions is improved, but computational cost and complexity increase

Engineering Contradiction:
Improveadaptability to unknown regionsVSAvoidcomputational complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent replaces machine learning techniques with a deterministic geometric optimization approach that uses convex optimization to generate lane graphs, eliminating the need for computationally expensive neural networks while maintaining adaptability to unknown regions

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system changes the approach from learning-based parameter estimation to deterministic geometric parameter optimization using convex optimization, which reduces computational complexity while preserving the ability to adapt to unknown regions through real-time sensor data

Inventive Principle:
Principle #35Parameter changes

3Speed

If approximate results from machine learning are accepted, then processing speed is improved, but measurement precision deteriorates

Engineering Contradiction:
Improveprocessing speedVSAvoidlane graph accuracy
Core Design Contradiction:
SpeedVSMeasurement precision

Solution Approach 1:

The patent replaces approximate machine learning inference with deterministic convex optimization that guarantees mathematically optimal solutions, achieving both high processing speed through efficient optimization algorithms and high measurement precision through guaranteed optimality

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentEP4160152A1Lane search for self-driving vehicles
Publication Date: 2023.04.05 WAYMO LLC
  • EP4160152A1 patent drawingFigure 1A~1D
  • EP4160152A1 patent drawingFigure 1E~1F
  • EP4160152A1 patent drawingFigure 1G~1J

AI summary

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium that create lane graph geometries from lane graph topologies. One of the methods includes receiving data representing a topological lane path through a plurality of cells of a drivable region. From the topological lane path, an initial polyline that traverses the same plurality of cells as the topological lane path can be generated. The initial polyline can be defined by vertices located on edges of a triangulated decomposition of the drivable region. A geometry optimization process can be performed on the initial polyline to generate a final polyline according to optimization criteria. From the final polyline, a geometric lane path representing a geometry of a drivable lane that traverses the drivable region can be generated.