Closed-Loop Perception Model Updates for Autonomous Vehicle Fleets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current computer-aided perception systems in autonomous driving face limitations due to sensor sensitivity to environmental conditions, high computational costs for depth estimation, and the need for manual data labeling, which hinders real-time adaptation and scalability in fleets of vehicles.
Innovation Solution
A system that automatically labels training data on-the-fly using cross modality and temporal validation, enabling continuous model updates and adaptation across a fleet of vehicles, with object-centric stereo for depth estimation and cross-validation to enhance accuracy and speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual or semi-manual labeling is used for training data, then data accuracy can be ensured, but the process becomes too slow and costly to adapt in real-time
Solution Approach 1:
The system performs automatic self-labeling of training data using its own detection models and cross-validation mechanisms, eliminating the need for manual human labeling while maintaining adaptability in real-time operation
Solution Approach 2:
The system implements closed-loop feedback where detection results are automatically validated, labeled, and used to update models in real-time, creating a continuous improvement cycle that maintains accuracy without manual intervention
2Measurement precision
If traditional stereo methods are used for depth estimation, then depth information can be obtained, but computational cost becomes too high and speed is insufficient
Solution Approach 1:
The system extracts only the essential depth information needed for detection tasks rather than computing complete depth maps, reducing computational burden while maintaining sufficient accuracy for autonomous driving applications
Solution Approach 2:
The system performs partial stereo processing focused on regions of interest rather than exhaustive processing of entire images, achieving adequate depth estimation speed for real-time operation
3Measurement precision
If 3D sensors such as LiDAR are used to acquire depth information, then depth data can be obtained, but the sensing range is limited and data density is low
Solution Approach 1:
The system merges data from multiple sensor types including 2D cameras and 3D sensors to compensate for individual sensor limitations, achieving extended effective range and improved data density through sensor fusion
4Adaptability or versatility
If 2D cameras are used for object detection, then the system can operate in various lighting conditions, but depth measurements are not provided
Solution Approach 1:
The system makes 2D camera data serve multiple functions by combining it with 3D sensor data to simultaneously achieve lighting robustness and depth estimation capability through integrated processing
Data Source
AI summary
The present teaching relates to method, system, medium, and implementation of a global model update center. At least one model is established at the model update center for detecting objects surrounding each of autonomous driving vehicles of a fleet. A plurality of labeled data items are received, from the fleet of autonomous driving vehicles, where each of the labeled data items is detected, based on the at least one model, from sensor data characterizing surroundings of the autonomous driving vehicles. The labeled data items are generated automatically on-the-fly by the autonomous driving vehicles. Based on the received labeled data items, at least some of the models are updated and model update information is accordingly generated. Such generated model update information is then distributed to the fleet of autonomous driving vehicles.