ADS Experience Monitoring With On-Board Anomaly Filtering

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current methods for monitoring and managing experiences of automated driving systems (ADS) in vehicles face challenges due to the impracticality of transferring large volumes of data over high bandwidth connections, which consume communication and power resources, and the inefficiency of sharing raw sensor data from a fleet of vehicles to a central server.

Innovation Solution

Implementing a machine learning model, such as an autoencoder, on-board each vehicle to identify new or rare experiences by exaggerating reconstruction errors, allowing only anomalous data to be transmitted to a remote server for analysis, thereby reducing the amount of data transferred and improving monitoring efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If all sensed data from the vehicle is transferred to the remote server for analysis, then complete monitoring of ADS experiences is achieved, but communication resources and power resources are excessively consumed

Engineering Contradiction:

ImproveADS monitoring reliabilityVSAvoidpower consumption for data transfer

Core Design Contradiction:

ReliabilityVSUse of energy by moving object

Solution Approach 1:

The patent extracts only the essential information (anomaly detection results and reconstructed data) from the complete sensed data stream. The autoencoder model processes all sensor data locally and extracts only the anomalous portions along with their reconstructions, which are then transmitted to the remote server. This extraction principle reduces the data volume for transmission while maintaining monitoring reliability.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent segments the data processing into two parts: local preprocessing using the autoencoder model on the vehicle, and remote analysis on the server. This segmentation allows the vehicle to filter and prepare data locally, transmitting only necessary segments (anomalies and reconstructions) rather than all raw data, thereby reducing communication and power resources.

Inventive Principle:

Principle #1Segmentation

2Reliability

If all sensed data from the vehicle is transferred to the remote server for analysis, then complete monitoring of ADS experiences is achieved, but communication bandwidth is excessively consumed

Engineering Contradiction:

ImproveADS monitoring reliabilityVSAvoiddata transfer volume

Core Design Contradiction:

ReliabilityVSQuantity of substance

Solution Approach 1:

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent uses the autoencoder to create a compressed representation (copy) of the anomalous data. Instead of transmitting the full high-dimensional sensor data, the system transmits the reconstructed version from the autoencoder, which captures the essential features of the anomaly in a more compact form, reducing bandwidth consumption.

Inventive Principle:

Principle #26Copying

3Quantity of substance

If a machine learning model is implemented on-board each vehicle to identify anomalous experiences, then data transfer volume is reduced, but device complexity increases

Engineering Contradiction:

Improvedata transfer volumeVSAvoidon-board processing complexity

Core Design Contradiction:

Quantity of substanceVSDevice complexity

Solution Approach 1:

The patent introduces the autoencoder model as an intermediary component between the sensor data and the transmission process. This intermediary processes the raw sensor data locally, identifying and reconstructing only the anomalous portions before transmission. The autoencoder acts as a mediator that simplifies the overall system by handling complex processing locally while maintaining simple transmission requirements.

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS12534103B2Methods and systems for automated driving system experience monitoring and/or management

Publication Date: 2026.01.27 ZENUITY AB

AI summary

Various methods for managing and/or monitoring new experiences of an Automated Driving System, ADS, of a vehicle are disclosed. In one example, an ADS experience library is used as training data for an autoencoder on-board the vehicle. A data stream representing driving experiences encountered by the vehicle as it is being driven around is suitably segmented and passed through the autoencoder. The output of the autoencoder exaggerates the reconstruction error for any data in a data segment not included in the training data set. This enables anomalous behavioural data indicative of a new experience encountered by a vehicle when it is being driven around to be identified in the data stream passed through the autoencoder, making it possible to send just the anomalous behavioural data to a back-end fleet server configured to monitor and manage a fleet of vehicles in a timely and bandwidth efficient manner.