Multi-Sensor RDA Detection for Autonomous Driving Trajectories

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing autonomous vehicle technologies struggle to accurately and reliably detect reduced drivability areas (RDAs) such as closed-off lanes and construction zones due to the challenges of collecting representative training data and relying on rule-based heuristics that result in false positives or missed detections.

Innovation Solution

An end-to-end (E2E) perception system that leverages multi-sensor modalities like cameras, radar, and lidar, combined with predictive learning, to efficiently generalize and detect RDAs using unlabeled data, providing accurate classification and navigation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If rule-based heuristics are used to detect reduced drivability areas, then the system is easier to implement, but detection accuracy deteriorates due to false positives and missed detections

Engineering Contradiction:
Improveease of implementationVSAvoiddetection accuracy
Core Design Contradiction:
Ease of manufactureVSMeasurement precision

Solution Approach 1:

The patent replaces rule-based heuristic methods with a deep learning-based perception system that uses neural networks to process sensor data. This substitution transitions from mechanical rule-based logic to an adaptive learning system, improving detection accuracy while maintaining implementation feasibility through modular architecture.

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

Solution Approach 2:

The system changes the fundamental parameter of detection from fixed rules to dynamic learned parameters. The neural network learns optimal detection parameters from training data, allowing the system to adapt to various RDA scenarios without requiring manual rule updates, thereby improving both accuracy and flexibility.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If representative training data is collected for RDA detection, then detection accuracy improves, but data collection complexity and time increase

Engineering Contradiction:
Improvedetection accuracyVSAvoiddata collection time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent employs a multi-sensor perception system that processes camera, radar, and lidar data through a unified neural network architecture. This universal system can detect various types of RDAs (construction zones, closed lanes, emergency scenes) using the same framework, reducing the need for separate data collection campaigns for each scenario.

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

Solution Approach 2:

The system performs preliminary data collection and model training offline before deployment. The neural network is pre-trained on comprehensive datasets that capture diverse RDA scenarios, allowing the system to be deployed with pre-acquired knowledge without requiring real-time data collection during operation.

Inventive Principle:
Principle #10Preliminary action

3Reliability

If multi-sensor modalities are integrated for RDA detection, then detection reliability improves, but system complexity increases

Engineering Contradiction:
Improvedetection reliabilityVSAvoidsystem complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent merges multiple sensor modalities (camera, radar, lidar) into a unified perception system. The neural network integrates features from all sensors to make joint predictions about RDAs, improving reliability through sensor fusion while managing complexity through a cohesive architectural framework.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The system segments the complex multi-sensor processing into distinct modular components: individual sensor data processing modules, feature extraction modules, and a final prediction module. This segmentation allows each component to be optimized independently while maintaining overall system reliability.

Inventive Principle:
Principle #1Segmentation

4Adaptability or versatility

If end-to-end predictive learning is used, then generalization ability improves, but computational requirements and processing time increase

Engineering Contradiction:
Improvegeneralization abilityVSAvoidprocessing time
Core Design Contradiction:
Adaptability or versatilityVSLoss of time

Solution Approach 1:

The system performs computationally intensive training and feature learning offline during a preliminary phase. The neural network is trained on extensive datasets to learn generalizable patterns, and this pre-learned knowledge is then applied during real-time operation, reducing online processing time while maintaining high generalization ability.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4624980A1End-to-end detection of reduced drivability areas in autonomous vehicle applications
Publication Date: 2025.10.01 WAYMO LLC
  • EP4624980A1 patent drawingFigure 1
  • EP4624980A1 patent drawingFigure 2
  • EP4624980A1 patent drawingFigure 3A

AI summary

The disclosed systems and techniques facilitate efficient detection and classification of traffic signs in driving environments. The disclosed techniques include, obtaining, using a sensing system of a vehicle, a set of camera images, a set of radar images, and a set of lidar images of an environment. The techniques further include generating, using a first neural network (NN), camera feature(s) characterizing the camera images, generating, using a second NN, radar features characterizing the radar images, and generating, using a third NN, lidar feature(s) characterizing the lidar images. The techniques further include processing the camera feature(s), the radar feature(s), and the lidar feature(s) to obtain an indication of a reduced drivability area in the environment.