Data-Driven Development
For the development and validation of applications based on artificial intelligence (AI), data-driven development is the method of choice in the industry today. It has decisive advantages because it is faster, simpler, and more efficient. AI is often used for it. The automotive industry in particular uses this method to develop automated or autonomous driving. This requires vast amounts of data from real sensors. Preparation and processing require time and resources. This is where we come in – because here, too, dSPACE offers support at every stage.
A data cycle takes place during the data-driven development of autonomous vehicles. This cycle includes data collection on the road, data acquisition with high bandwidth in a data center, analysis of the data for relevant traffic situations, and labeling for AI training and for generating ground truth in data replay tests.
Data Logging – Making Data Available Faster and More Cost-Effectively
If we think of data-driven development as a data pipeline, then data logging can be considered an elementary part of it. It is the starting point of the pipeline, and is followed by further development steps such as data acquisition, data analysis, data selection, data commenting and, of course, software development, accompanied by comprehensive testing and validation, including data replay tests.
dSPACE offers a wide range of data acquisition hardware that covers all kinds of data acquisition requirements, from single sensor applications to complete data acquisition fleets. It also supports relevant automotive buses, networks, and sensor interfaces and is highly scalable.
Data Ingestion
Data recording is the next step in the data pipeline after the data has been recorded in the vehicle. Data ingestion is about bringing dozens of terabytes that are collected daily in every vehicle in the fleet into the final repository and making them accessible to users.
dSPACE offers two dedicated AUTERA Upload Stations to use the data on a developer's PC or to upload the recorded data to an existing server infrastructure or to the cloud as quickly as possible. The AUTERA Upload Station for uploading data to a server infrastructure can read up to two AUTERA SSDs simultaneously and stream the data directly to the data center, for example, via 100 Gigabit Ethernet
Data Annotation and Anonymization in First-Class Quality
Not all data is equally valuable. When it comes to training AI systems, quantity alone is not enough. Selecting the right data from your data lake is crucial for training algorithms to be able to handle rare but potentially critical situations and scenarios.
Intempora, a dSPACE company, offers the Intempora
Validation Suite (IVS), a cloud-based data manager with a test automation framework that provides unique methods and workflows to help you bring autonomous vehicles onto the road faster than ever before. In addition, the anonymization of faces and license plates in recorded data has become a global requirement. dSPACE, with its company understand.ai, offers the right solutions to meet these high demands.
Sensor Data Management with the Right Tools
Cost-effective processing of the collected sensor data remains a major challenge for autonomous driving. Separate frameworks, different IT infrastructures, and different data formats pose major challenges today and the wrong cloud strategy can cause costs to explode.
IVS combines manual and automatic indexing modules with AI-based scene understanding to build a catalog of your sensor data, allowing end users to archive inefficient data segments and leverage useful ones, speeding up the development process and reducing data storage and management costs.
dSPACE Data Management Tooling also offers the possibility to combine the scalability benefits of public clouds with the data protection and cost benefits of on-premise clusters by allowing hybrid cloud implementations to combine the advantages of public clouds and on-premise providers.
Powerful data management helps centralize, store, search, filter, preview, post-process, and analyze recorded data for the data-driven development of functions for ADAS/automated driving and other mobility applications. With the dSPACE solution for sensor data management, you can quickly find what you are looking for in petabytes of traffic data.
Training Artificial Intelligence
Which data is the right data? For example, the AI has to learn what a person looks like.
Intempora RTMaps provides a framework for data acquisition, real-time data processing, perception algorithm development, sensor fusion, and trajectory planning. RTMaps offers a precise time-stamping mechanism for data logging, replay & fusion, easy integration of custom C++, Python, and Simulink® code, quick start and ready-to-use components, offline visualization and integration options, replay control, 2D image, bird’s-eye and 3D viewers, pre-built sensor interfaces, and customizable layouts.
Scenario Generation
In order to ensure the safe operation of automated driving functions in all conceivable situations and ultimately to prove this for approval, the vehicle must be extensively tested. As it is unrealistic in terms of time and costs to drive several million test kilometers with a real vehicle to provide this proof, simulation is used.
Meaningful simulation scenarios represent relevant situations in the vehicle environment.
These scenarios can be generated by processing recorded data from real test drives, among other things. The dSPACE tools process raw data from vehicle sensors, object lists, or other data sets and generate simulation scenarios that are even suitable for physical sensor simulation in a highly automated process. AI-based methods and clever algorithms are used to classify certain types of driving maneuvers in the recorded measurement data.
With the help of this process, critical situations that occur during test drives can be repeatedly simulated and taken into account in regression tests. The flexible parameters of the generated simulation scenarios also make it possible to turn less critical real-life situations into critical situations in the simulation, for example, by changing speeds or distances, and thus place additional demands on the function for automated driving. The consistent application of this procedure throughout the entire development and validation process leads to the creation of a large database of relevant simulation scenarios over time, which forms a solid basis for the vehicle manufacturers' safety arguments.
Data Playback
While simulations can create critical traffic situations that are difficult to reproduce in reality, the accuracy of simulation environments is always questionable, as comprehensive modeling of all real-world factors is difficult to achieve.
Therefore, data from the real world is needed and used to compensate for deficiencies in the simulation and to cover as many of the different conceivable traffic scenarios as possible within the operational design range of the developed algorithm.
The dSPACE data replay solution gives you the tools to replay recorded data, whether it is real or synthetic, while ensuring accurate synchronization of sensor and network/bus data streams for the system under test.
The accompanying test and data management software enables easy test scaling and lets you cover millions of test kilometers with virtual and real ADAS/AD components.
