Safety Validation Solutions for Automotive Perception Functions
aidkit is a ML Quality Ops platform built for ML Engineers and designed for their workflow.
Its features allow users to systematically ensure every safety-relevant aspect is covered, tested, and evidenced prior to deployment of a ML-augmented perception component.
Safety Validation Solutions for Automotive Perception Functions
aidkit is a ML Quality Ops platform built for ML Engineers and designed for their workflow.
aidkit’s features allow users to systematically ensure every safety-relevant aspect is covered, tested, and evidenced prior to deployment of a ML-augmented perception component.
Safety assurance with aidkit
Validate Perception Function Safety across ODDs and Scenes before Deployment
Model is detecting or classifying correctly
Model is failing to detect or classifying incorrectly
Training Data vs. Real World
Unexpected attributes in the real-world environment can lead a model to malfunction. Testing models in challenging scenarios before they occur in production prevents severe consequences.
Advanced Feature-set
aidkit ’til You Make It
Extract risk scores for safety argumentation
Provide explainable, evidence-based risk scores containing impact and probability of occurrence for every tested ODD scenario
Provide examples that challenge your perception component
Generate a wide range of corrupted images for average- and worst-case analysis (i.e., perturbation-based image operations)
Discover and explore new ODD scenarios
Combine existent ODD attributes to automatically identify ODD extensions, interaction effects, and correlations
Tailor your Data Campaign and augment training & test data
Know exactly what data to collect and label to meet your performance shortfalls and leverage the corrupted images your model could not handle for re-training
Track data lineage and fulfill audit requirements
Maintain an overview of existing models, model versions, data sets/splits ,and evaluations plus monitor interim pipeline results
Automate and accelerate perception component testing
Connect aidkit to your existing MLOps workflow via our Python-Client and scale your computing to execute testing faster
Your aidkit Benefits
aidkit Covers every Aspect for Testing your Perception System
Identify and test all relevant scenarios, whether known or unknown
aidkit offers easy testing of diverse scenarios, as well as the identification of new safe and unsafe areas via the combination of existing ODD attributes
Discover what data you need to improve and augment your data with artifacts
aidkit identifies the edge and corner cases where you need more data so your data collection and labelling is targeted, or import artifacts directly for retraining
Evidence for Safety
aidkit generates artifacts backed by empirical evidence and risk scores for each scenario allowing easy comparison and support for safety claims
Up-to-date on the latest standards and the means to test them
aidkit updates benefit from neurocat’s active engagement in developing the latest industry standards (e.g. ISO PAS 8800) which we integrate as state-of-the -art testing methods
Track, automate, scale
aidkit facilitates easy tracking of data lineage and metrics, monitoring of pipelines including interim results, automate iterative testing and retraining, and horizontal scaling
One Platform with two flavours for all your testing needs
Input- and framework-agnostic tooling facilitates testing all your use-cases in a single platform available as a REST-API you can integrate right in your workflow and a Web interface for configuration and tracking
Customized corruptions
neurocat will work with you on the approximation of your ODD catalogue with customized augmentation
Support for every iteration
neurocat is there to support you with the tailored solution projects you need, from aidkit integration to technical challenges of safety validation to results interpretation
