We Generate Ground Truth Annotations

We are the autonomous driving experts for ground truth data annotation. From 3D LiDAR Point Cloud Annotations to 2D Bounding Boxes or 3D Sensor Fusion Annotation – understand.ai covers the broad diversity of all regular sensor data formats and annotation types. Through our end-to-end services and state-of-the art DevOps, we make your training and validation projects feasible and ensure high quality for ground truth annotations!


To examine whether the sensors and perception models of autonomous and ADAS systems work correctly under real world conditions, validation projects are a necessary requirement to ensure everyone's safety on the road.

In order to run such a validation project successfully, large amounts of high quality validation data with high recall rates are required. In addition, the dataset needs to be diverse and representative - covering everything from standard scenarios to edge cases and critical situations.

Thanks to high levels of automation, we can deliver more ground truth annotations in a shorter time accelerating your validation projects. No matter what part of the automated driving stack is subject to testing and validation, we have your project covered.


In assisted and autonomous driving, perception algorithms are used to train autonomous driving functions in recognizing their environment.

Having ground truth data annotations during the training process is imperative to correctly learn the target data domain. Imprecise annotations and lack of diversity in the training dataset have a detrimental effect on the learning process. Through our end-to-end-services, we not only accelerate your training projects but get the ground truth quality right.

Scenario Generation

It’s not always feasible to test and validate your autonomous driving functions with real-world driving tests. We help you cover the enormous test space through highly accurate simulations.

Through the process of Scenario Identification, we identify relevant scenarios derived from recorded real-world data. To create precise replay scenarios, we extract and localize all objects along with their class and trajectory during Scenario Extraction and create the necessary variations through the process of Scenario Fuzzing. Challenge your driving functions with complex scenario generation and eliminate bugs early in the development process.