Neural Network Training via Vehicle Perception and Communication
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The labor-intensive process of generating and validating large training data sets for neural networks, particularly in autonomous vehicles, is inefficient and accuracy-dependent on manual classification, which is a critical concern for reliable obstacle detection and classification.
Innovation Solution
A method and system for automatically training neural networks using perception recorders and communication devices that continuously record surroundings and receive positional and type information from objects, enabling real-time identification and classification of objects for training purposes within a connected vehicle ecosystem.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If manual tagging and classification of training data is performed, then the training data can be created and labeled, but the process becomes labor-intensive and time-consuming
Solution Approach 1:
The system enables vehicles to automatically generate and tag their own training data by recording sensor information and correlating it with communication messages from other vehicles. The vehicle's own perception recorder and communication device work together to create labeled training data without external manual intervention, allowing the system to self-generate training datasets at scale
Solution Approach 2:
The system pre-correlates sensor data with communication messages during normal vehicle operation to create pre-tagged training data. By performing the classification and tagging actions in advance during data collection, the system eliminates the need for subsequent manual labeling efforts when training datasets are needed
2Measurement precision
If manual classification of objects in images is performed, then training data can be labeled, but the accuracy depends on the classifier's reliability which is a matter of trust
Solution Approach 1:
The system uses communication messages from other vehicles as an intermediary to provide ground truth object information. Instead of relying on manual classification or untrusted automated classifiers, the system correlates sensor data with authenticated communication messages from vehicles that directly observe and report object positions and types, creating a trusted classification mechanism
Solution Approach 2:
The system uses communication messages as feedback to verify and validate the classification of objects detected by perception sensors. The received messages provide independent verification of object presence, type, and position, allowing the system to cross-check and confirm classifications with high reliability
3Quantity of substance
If large training data sets are generated and manually tagged, then the neural network can be trained, but the process is inefficient and labor-intensive
Solution Approach 1:
Vehicles automatically generate and tag their own training data using their perception recorders and communication devices during normal operation. This self-service approach eliminates the need for external manual tagging operations, allowing unlimited training data generation without proportional increases in human labor
Solution Approach 2:
The system continuously generates training data during normal vehicle operation by constantly recording sensor information and correlating it with incoming communication messages. This continuous data generation process transforms idle operational time into productive training data creation, significantly increasing overall productivity
Data Source
AI summary
A method for automatically training a neural network includes at a trainer having a first communication device and a perception recorder, continuously recording the surroundings in the vicinity of the first object; receiving, at the trainer, a message from a communication device associated with an object in the vicinity of the trainer, the message including information about the position and the type of the object; identifying a recording corresponding to the time at which the message is received from the object; correlating the received positional information about the second object with a corresponding location in the recording to identify the object in the recording; classifying the identified object based on the type of information received in the message from the object; and using the classified recording to train the neural network.