Autonomous Vehicle Sensor Fusion Positioning With GPU Constraint Processing

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Positioning calculations using visual/LiDAR odometry typically consume a large amount of CPU resources, necessitating the development of an odometry system based on vision, LiDAR, and inertial measurement units that is both precise and resource-efficient.

Innovation Solution

A method is introduced that acquires sensor data from autonomous vehicles, including inertial measurement data, LiDAR data, and visual image data, and processes it using constraint factors to obtain positioning data, leveraging GPU resources for efficient processing and reducing CPU usage.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If visual/LiDAR odometry is used for positioning calculations, then positioning accuracy is improved, but CPU resource consumption increases

Engineering Contradiction:
Improvepositioning accuracyVSAvoidCPU resource consumption
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent segments the computing tasks by separating visual processing and LiDAR processing onto different GPUs, and further divides the computation into parallel threads. Each GPU handles specific sensor data independently, then results are merged to achieve accurate positioning without overloading the CPU.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces an intermediary processing layer that uses constraint factors to bridge visual and LiDAR data. This intermediary approach allows the system to process multiple sensor types efficiently by establishing mathematical relationships between them, reducing the computational burden on the CPU while maintaining positioning accuracy.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Measurement precision

If multiple sensors (visual, LiDAR, IMU) are processed together, then positioning accuracy is improved, but device complexity increases

Engineering Contradiction:
Improvepositioning accuracyVSAvoidprocessing system complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent merges visual, LiDAR, and IMU data processing into a unified framework using constraint factors. By combining multiple sensor inputs through a common mathematical model, the system achieves high positioning accuracy while managing complexity through integrated processing rather than separate independent systems.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent creates a universal processing framework that handles multiple sensor types (visual, LiDAR, IMU) through a common constraint factor approach. This multi-functional system can process different sensor data types using the same underlying methodology, reducing overall system complexity despite handling diverse inputs.

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

3Speed

If GPU resources are utilized for sensor data processing, then computing speed is improved, but resource allocation complexity increases

Engineering Contradiction:
Improvecomputing speedVSAvoidresource allocation complexity
Core Design Contradiction:
SpeedVSDevice complexity

Solution Approach 1:

The patent transitions from single-CPU processing to a multi-dimensional parallel processing architecture using multiple GPUs. By moving computation to the GPU dimension and utilizing parallel thread execution, the system achieves faster computing speed while managing allocation complexity through structured thread organization and independent GPU operation.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Data Source

PatentEP4160271B1Method and apparatus for processing data for autonomous vehicle, electronic device, and storage medium
Publication Date: 2024.05.01 APOLLO INTELLIGENT DRIVING (BEIJING) TECHNOLOGY CO LTD
  • EP4160271B1 patent drawingFigure 1
  • EP4160271B1 patent drawingFigure 2
  • EP4160271B1 patent drawingFigure 3~4

AI summary

A method and apparatus for processing data for an autonomous vehicle, an electronic device, a storage medium, and a computer program product are provided, which relate to a field of an intelligent transportation technology, in particular to a field of autonomous driving. The method of processing data for the autonomous vehicle includes: acquiring sensor data for the autonomous vehicle, wherein the sensor data includes inertial measurement data, LiDAR data, and visual image data; determining a first constraint factor for the inertial measurement data according to the inertial measurement data and the visual image data; determining a second constraint factor for the LiDAR data according to the inertial measurement data and the LiDAR data; determining a third constraint factor for the visual image data according to the inertial measurement data, the visual image data and the LiDAR data; and processing the sensor data based on the first constraint factor, the second constraint factor and the third constraint factor, so as to obtain positioning data for positioning the autonomous vehicle.