Autonomous Vehicle Perception via Multi-Pipeline Point Cloud Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional systems for autonomous vehicles sacrifice computational speed and accuracy in favor of higher resolution data, making it challenging to identify objects and their movements within the scene, particularly for navigational actions.
Innovation Solution
The system generates multiple perceptions of a scene using a point cloud, including object location, speed, classification, and type, by projecting the point cloud onto a horizontal plane and using deep learning, and selects the most accurate and high-resolution perception for motion planning, combining data from RADAR and LIDAR sensors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If higher resolution data is used for scene perception, then measurement precision is improved, but computational speed deteriorates
Solution Approach 1:
The patent segments the scene perception task into multiple independent perception pipelines (e.g., occupancy grid pipeline, baseline pipeline, deep learning pipeline), each processing different aspects of the scene at different resolutions. This allows parallel processing of multiple perceptions simultaneously, maintaining high accuracy while improving computational throughput by dividing the overall task into manageable segments that can be executed concurrently.
2Measurement precision
If higher resolution data is used for scene perception, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The system is divided into multiple specialized perception pipelines, each with a specific function and resolution level. This segmentation allows the complex overall task to be managed through simpler, focused sub-tasks that can be independently optimized and maintained, reducing the effective complexity burden on any single component while achieving high overall precision.
Solution Approach 2:
The patent implements a universal perception framework where multiple pipelines process the same input data (point cloud) but produce different types of perceptions (occupancy grids, object classifications, semantic segmentations). This multi-functional architecture allows a single system to handle diverse perception requirements without duplicating hardware or data acquisition systems, managing complexity through software-based functional differentiation.
3Reliability
If multiple perception pipelines are executed, then reliability is improved, but productivity decreases
Solution Approach 1:
By segmenting the perception task into parallel pipelines, the system achieves reliability through redundancy and cross-validation of multiple independent analyses. Each pipeline contributes a different type of perception that can be independently verified, improving reliability without requiring sequential execution, as the segmented pipelines operate simultaneously to maintain high processing throughput.
Solution Approach 2:
The patent maintains continuous perception by having multiple pipelines operate in parallel rather than sequentially, ensuring that useful computational action is continuously performed across all pipelines simultaneously. This continuous parallel processing maintains high productivity while the coordinated output from multiple continuous pipelines enhances reliability through consistent, multi-faceted scene understanding.
Data Source
AI summary
Systems and methods for perceiving a scene around a mobile device include a sensor system capturing an image of and object list data for the scene around the mobile device. The image includes a point cloud with group(s) of points. An object list is generated from the object list data, and the object list indicates a location and a speed of each surface in a set of surfaces in the scene. A perception system generates a selected perception using the image and the object list. The selected perception is generated based on a comparison of first, second, and third perceptions within a hierarchy. The first perception includes object speed and object location. The second perception includes object classification and object tracking, and the third perception includes object type and object model. A motion planner generates a motion plan for the mobile device based on the selected perception.