Directional Control Data Likelihood Estimation for Anomaly Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing computer-controlled systems face challenges in accurately handling unexpected situations due to insufficient training data, leading to potential harmful actions, and there is a need for improved methods to estimate reliability of control data and generate synthetic data representative of training datasets, especially for directional data.

Innovation Solution

A generative model using a coupling layer that applies rotations and circle transformations to direction vectors, allowing for accurate estimation of reliability and generation of synthetic directional data, which can be efficiently evaluated on parallel processors.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If a trained computer-controlled system is deployed in practice, then the system can perform automated control tasks, but it may encounter unexpected situations that it did not encounter during training, leading to harmful actions

Engineering Contradiction:
Improveautomated control capabilityVSAvoidsafety in unexpected situations
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The system performs preliminary actions by training anomaly detection models during the training phase on the full set of operating conditions. These models are prepared in advance to detect unexpected situations before the system is deployed, allowing the system to identify and handle anomalies without harmful actions occurring during actual operation.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

An anomaly detection model serves as an intermediary between the trained control system and the real-world environment. This intermediary layer analyzes incoming data and determines whether the current situation conforms to training data distributions, enabling the system to recognize unexpected conditions and trigger appropriate safety responses before harmful actions can occur.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Reliability

If anomaly detection is used to detect unexpected situations, then the system can identify non-conforming states, but the robot may be halted or switched to safe mode, reducing operational efficiency

Engineering Contradiction:
Improveanomaly detection accuracyVSAvoidoperational efficiency
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The system applies different response strategies based on the local quality of the anomaly detection result. Instead of uniformly halting operation for all detected anomalies, the system can selectively respond based on the severity and type of anomaly, allowing normal operation to continue for minor deviations while triggering safety modes only for critical anomalies, thus maintaining operational efficiency while ensuring safety.

Inventive Principle:
Principle #3Local quality

3Adaptability or versatility

If deep neural networks are trained end-to-end for control tasks, then the system can learn complex control policies, but the training data may not be representative of all operating conditions, leading to poor generalization

Engineering Contradiction:
Improvecontrol policy learning capabilityVSAvoidgeneralization to unseen conditions
Core Design Contradiction:
Adaptability or versatilityVSReliability

Solution Approach 1:

The system implements feedback by using anomaly detection models to monitor whether the current operating conditions conform to the training data distribution. This feedback mechanism allows the system to identify when it is operating in unseen conditions and can trigger appropriate responses such as halting operation or switching to safe mode, thereby improving reliability and generalization performance without compromising the learned control policies.

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP3975055B1Estimating reliability of control data
Publication Date: 2025.09.24 ROBERT BOSCH GMBH
  • EP3975055B1 patent drawingFigure 1~2
  • EP3975055B1 patent drawingFigure 3
  • EP3975055B1 patent drawingFigure 4

AI summary

The invention relates to a computer-implemented method (600) of estimating a reliability of control data for a computer-controlled system interacting with an environment. The control data is inferred from a model input by a machine learnable control model which is trained on a training dataset. The model input comprises at least one direction vector which is extracted from sensor data and which is associated with a component of the computer-controlled system or an object in the environment. The reliability is estimated using a generative model that is trained to generate synthetic model inputs representative of the training dataset, by applying an inverse of the generative model to the model input to determine a likelihood of the model input being generated according to the generative model. The generative model comprises a coupling layer comprising a circle transformation and one or more of an unconditional rotation and a conditional rotation.