Vehicle Perception Pipeline Using Out-of-Sequence Radar Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional perception systems for autonomous vehicles face significant processing delays and latency in detecting, classifying, and predicting the motion of objects, which can compromise safety, especially when objects change trajectory rapidly.
Innovation Solution
The out of sequence perception system implements a fast object tracking update pipeline that uses intermediate radar data to generate more recent object data, reducing latency by processing sensor data more frequently than traditional systems, and combines this with a full object tracking update pipeline to maintain accuracy and recency of sensor data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional perception systems process sensor data through complete detection, classification, and prediction pipelines, then comprehensive object information is obtained, but processing time and latency increase significantly
Solution Approach 1:
The perception system is divided into two independent pipelines: a full object tracking update pipeline that processes all sensor data types (lidar, radar, video) for comprehensive object information, and a fast object tracking update pipeline that processes only radar data for rapid updates. This segmentation allows the system to obtain comprehensive object information through the full pipeline while simultaneously maintaining low-latency updates through the fast pipeline, resolving the contradiction between detection accuracy and processing latency.
Solution Approach 2:
The fast object tracking update pipeline performs preliminary processing of radar data to generate rapid object position and velocity updates. By preprocessing radar data independently and feeding results into the full pipeline, the system prepares object tracking information in advance, reducing the overall processing time required for comprehensive object detection and classification.
2Productivity
If sensor data is processed at lower frequencies to reduce computational load, then processing speed increases, but the recency and accuracy of object tracking data decrease
Solution Approach 1:
The system implements dynamic processing frequencies for different data types: radar data is processed at high frequencies (e.g., 500 Hz) to maintain current object position and velocity information, while lidar and video data are processed at lower frequencies (e.g., 10-30 Hz) for comprehensive classification. This dynamic frequency adjustment allows the system to maintain high processing throughput while ensuring object tracking accuracy through frequent radar updates.
Solution Approach 2:
Radar data serves as an intermediary that bridges the gap between high-frequency processing requirements and lower-frequency comprehensive sensor processing. The fast pipeline uses radar data to maintain up-to-date object tracking information, which then informs and supplements the full pipeline processing of lidar and video data, ensuring both throughput and reliability.
3Loss of information
If complete sensor fusion pipelines are used for all objects, then comprehensive object classification is achieved, but processing time increases for rapidly changing objects
Solution Approach 1:
The system applies different processing qualities to different objects based on their characteristics: high-speed or dynamically changing objects receive frequent updates from the fast radar-based pipeline that prioritizes position and velocity tracking, while stationary or slow-moving objects are updated by the full pipeline at standard intervals. This local quality differentiation ensures rapid response for dynamic objects without sacrificing classification completeness for all objects.
Solution Approach 2:
For rapidly changing objects, the system performs partial processing through the fast pipeline that focuses on critical parameters (position, velocity) using radar data, rather than waiting for complete processing of all sensor data types. This partial action provides timely response for dynamic objects, while the full pipeline continues to process complete sensor fusion data for comprehensive classification when time permits.
Data Source
AI summary
Techniques for updating data operations in a perception system are discussed herein. A vehicle may use a perception system to capture data about an environment proximate to the vehicle. The perception system may receive image data, lidar data, and/or radar data to determine information about an object in the environment. As different sensors may be associated with different time periods for capturing and/or processing operations, the techniques include updating object data with data from sensors associated with a shorter time period to generate intermediate object data.