Neural-Network Driver Assistance for Continuous Obstacle-Aware Pathing

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional driver-assistance systems face challenges in predicting vehicle paths due to the presence of static and dynamic objects on the road, as they require precise target positions and stepwise trajectories, which can be laborious and inexact.

Innovation Solution

A deep neural network is used to create a continuous mathematical function representing a path for the vehicle, which is updated regularly, allowing the vehicle to circumnavigate obstacles optimally without relying on target positions, utilizing a grid populated with object detection information and incorporating convolutional layers and polynomial coefficients for path prediction.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional trajectory calculation methods are used that rely on known target positions, then the path prediction can be calculated stepwise, but the process becomes laborious and inexact when objects are merely detected without target position information

Engineering Contradiction:
Improvepath prediction accuracyVSAvoidcalculation process complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent replaces conventional mechanical trajectory calculation methods with a neural network-based system. Instead of using stepwise trajectory calculations that require known target positions, the neural network directly processes detected objects and their attributes to predict the vehicle path in a continuous manner, eliminating the need for manual target position determination and reducing calculation complexity

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

Solution Approach 2:

The patent changes the input parameters from requiring known target positions to using only detected object attributes (type, position, velocity, acceleration). The neural network learns to map these parameters directly to path predictions, transforming the problem from a constraint-based calculation to a data-driven prediction that works with available sensor data

Inventive Principle:
Principle #35Parameter changes

2Adaptability or versatility

If stepwise trajectories with multiple target positions are calculated to account for different object behaviors, then the path prediction can handle static and dynamic objects, but the calculation becomes more laborious and requires several target positions

Engineering Contradiction:
Improvehandling of static and dynamic objectsVSAvoidnumber of target positions required
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent creates a universal neural network model that handles both static and dynamic objects through a single integrated architecture. The network processes all detected objects uniformly, using their attributes (type, position, velocity, acceleration) to automatically adapt the path prediction, eliminating the need for separate trajectory calculations for different object types

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent uses a grid-based representation that copies the spatial environment into a discrete format, where each cell contains information about objects and their attributes. This grid representation serves as a simplified model that the neural network processes to predict the continuous path, replacing complex multi-target trajectory calculations with a unified grid-to-path mapping

Inventive Principle:
Principle #26Copying

3Productivity

If conventional systems require target positions for trajectory calculation, then the path can be predicted stepwise, but manual determination of target positions is laborious and inexact

Engineering Contradiction:
Improvepath prediction efficiencyVSAvoidtarget position accuracy
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The neural network system is self-sufficient in determining the path without requiring external input of target positions. It automatically processes detected object data, learns the mapping from object attributes to path predictions during training, and generates continuous path predictions independently, eliminating manual target position determination entirely

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP4250183A1Driver-assistance system with neural network
Publication Date: 2023.09.27 AUMOVIO AUTONOMOUS MOBILITY GERMANY GMBH
  • EP4250183A1 patent drawingFigure 1~2
  • EP4250183A1 patent drawingFigure 3a
  • EP4250183A1 patent drawingFigure 3b

AI summary

The invention relates to a driver-assistance system (100) for a vehicle (110) comprising a processing unit (102) configured to receive information about objects on a road, map the objects (120, 122) into a grid, create a mathematical function based on the grid, wherein the mathematical function represents a path (402) for the vehicle (110) on the road, and guide the vehicle (110) along the path (402).