Functional Road Map Generation From Onboard Vision for Autonomous Navigation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current autonomous vehicle navigation systems rely on pre-existing high-definition maps that require continuous updates and manual effort, and are not feasible without GPS connections, limiting their scalability and cost-effectiveness.
Innovation Solution
A method using real-time visual input from sensors to generate functional top-down road maps without pre-determined precise mapping, employing an encoder-decoder neural network architecture that transforms front-view images into top-down maps with relevant features for autonomous navigation, such as drivable roads and lane markings, without relying on GPS or pre-existing maps.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If pre-existing high-definition maps are used for autonomous navigation, then navigation accuracy is improved, but system cost and maintenance complexity increase
Solution Approach 1:
The autonomous vehicle generates its own functional maps in real-time using onboard sensors and neural networks, eliminating the need for external HD map infrastructure. The vehicle serves itself by creating navigation-relevant map data on-demand, transforming from a system that consumes pre-made maps to one that produces its own operational maps.
Solution Approach 2:
Instead of using expensive, pre-produced HD maps, the system creates simplified functional map copies that contain only the navigation-relevant features needed for autonomous driving. These functional maps are generated by processing sensor data through neural networks that extract and represent only essential road elements.
2Measurement precision
If pre-determined precise mapping is used, then map accuracy is improved, but adaptability to new locations deteriorates
Solution Approach 1:
The mapping system transitions from static pre-determined maps to dynamic real-time generation. The neural network continuously processes current sensor data to generate up-to-date functional maps adapted to the vehicle's current location and environment, enabling immediate adaptability to new locations without pre-existing map data.
Solution Approach 2:
The vehicle independently generates location-specific functional maps on-demand using its onboard sensors and processing systems, eliminating dependence on pre-surveyed map data for new locations. This self-service capability enables immediate operation in previously unmapped areas.
3Reliability
If manual HD map updates are performed, then map currency is improved, but productivity and cost efficiency deteriorate
Solution Approach 1:
The system automatically generates current functional maps in real-time using onboard sensors and neural networks, eliminating the need for manual map updating processes. The vehicle continuously creates its own updated map data, ensuring currency without human intervention or external infrastructure support.
Solution Approach 2:
The functional map generation operates continuously in real-time as the vehicle moves, constantly updating the map representation based on current sensor input. This continuous generation process replaces discrete manual update cycles, ensuring map currency without interruption to vehicle operation.
4Measurement precision
If GPS connection is required for map-based navigation, then location precision is improved, but system versatility and scalability deteriorate
Solution Approach 1:
The system extracts and removes the GPS dependency from the autonomous navigation architecture. By generating functional maps from onboard sensor data alone, the system separates navigation capability from external positioning infrastructure, enabling operation in locations without GPS coverage or connectivity.
Solution Approach 2:
The vehicle uses its own onboard sensors and processing systems to generate functional maps independently, without relying on external GPS infrastructure or cloud-based map services. This self-sufficient approach enables deployment in remote areas and eliminates scalability limitations imposed by GPS availability.
Data Source
AI summary
A method, an apparatus and a computer program for real-time generation of functional road maps. The method comprises obtaining a real-time input from a sensor mounted on a vehicle, that captures a front view of a road ahead of the vehicle and processing thereof by a neural network to generate a functional map of the road ahead of the vehicle. Each pixel in the functional map is associated with a predetermined relative position to the vehicle. A content of each pixel is assigned a set of values, each of which represents a functional feature relating to a location at a corresponding predetermined relative position to the pixel. The processing is performed without relying on a pre-determined precise mapping. The method further comprises providing the functional map to an autonomous navigation system of the vehicle, to autonomously drive the vehicle in accordance with functional features represented by the functional map.