On-Board Autoencoder Compression for Vehicle Sensor Storage
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for autonomous vehicle mapping face challenges in efficiently compressing and storing sensor data from vehicles, leading to bandwidth inefficiencies and limited on-board storage capacity.
Innovation Solution
The implementation of an autoencoder on-board vehicles to compress sensor data from image, LIDAR, and radar sensors, automatically discarding unnecessary data and preserving information required for scene reconstruction and understanding tasks, allowing for real-time storage and efficient bandwidth usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If lossless compression techniques are used to store sensor data, then data integrity is preserved, but storage capacity is rapidly exhausted and bandwidth efficiency deteriorates
Solution Approach 1:
The patent transforms sensor data from raw format to compressed representation by changing its parameter state through autoencoder encoding. The encoder network converts high-dimensional sensor data into lower-dimensional latent representations, achieving lossy compression that preserves essential features while discarding redundant information, thus resolving the contradiction between data integrity and storage capacity
Solution Approach 2:
The autoencoder architecture extracts and retains only the most important features of sensor data through its encoding process. The encoder learns to identify and preserve critical scene understanding information while eliminating unnecessary data, effectively separating essential from non-essential components to optimize storage efficiency
2Reliability
If all sensor data is uploaded to backend computing systems, then complete scene reconstruction is possible, but bandwidth consumption increases significantly
Solution Approach 1:
The system performs preliminary compression and feature extraction on-board the vehicle using the autoencoder before data transmission. By pre-processing the sensor data to extract essential features and discard redundant information, the system reduces the data volume requiring bandwidth transmission while ensuring that critical information for scene reconstruction is preserved
Solution Approach 2:
The autoencoder transforms sensor data parameters from raw high-dimensional format to compressed low-dimensional latent space representation. This parameter transformation maintains the essential characteristics needed for scene reconstruction while significantly reducing the data size that must be transmitted over the network
3Quantity of substance
If compression techniques discard data, then storage efficiency improves, but information needed for autonomous driving tasks may be lost
Solution Approach 1:
The autoencoder is trained in a self-supervised manner where the decoder learns to reconstruct the input data from the compressed representation. This self-service training process automatically teaches the system to preserve only the most important information needed for downstream tasks like scene understanding and object detection, while discarding redundant data
Solution Approach 2:
The reconstruction loss function provides feedback during training to guide the encoder in preserving essential information. The decoder's ability to reconstruct the input data serves as a feedback mechanism that ensures critical scene understanding information is retained in the compressed representation while allowing non-essential data to be discarded
Data Source
AI summary
A vehicle computing system can receive sensor data from a sensor suite of the vehicle. The system can execute a autoencoder on the sensor data to compress the sensor data, and can stored the compressed sensor data in an on-board database of the vehicle.