• Reuse of existing custom replay platforms in a cloud-based environment
• Data replay to software components and virtual processing units of any complexity
• One central tool for high-scale testing and comprehensive evaluation in the Cloud

Task

Perception and data fusion algorithms are predominantly based on computer vision and artificial neural networks, which can quickly reach a high level of complexity and therefore need to be tested for correct behavior at an early stage, parallel to their development and training with real data. Test developers often use their own solutions for this at their workplace or in the local test environment, which should be transferred to the cloud in later development phases without significant adjustments in order to achieve efficient validation across the board. Other test departments, on the other hand, are specifically looking for replay platforms that they can use consistently across all integration stages, regardless of whether they incorporate their perception function into simple software components or virtual ECUs. For all of them, there is an equal need to be able to easily visualize test results in the form of key performance indicators (KPIs), evaluate them conveniently, and compare them with previously collected data.  

Challenge

Safeguarding using software data replay can be carried out in all phases of the development of an ADAS/AD application at different levels of maturity. There are already a variety of self-developed and commercial solutions on the market that are successfully used in CI/CD pipelines parallel to application development. In order to achieve high scalability and make optimum use of global test capacities, more and more test departments are aiming to move a large proportion of validation to a global cloud. The central challenge here is to transfer the very heterogeneous test solutions to a cloud-based test framework without great effort and then to continue to use the advantages of 24/7 coverage in a proven manner. 

Solution

With abstract execution for the web-based test management tool SIMPHERA, dSPACE is setting a new standard for flexible data replay in the cloud. The abstract execution container defines an open and standardized interface for easy integration of replay platforms of any kind that can be executed in Docker. This allows individual replay solutions that have already been used successfully to be integrated into SIMPHERA data replay testing (DRT) with little effort, with advantages such as easy use of global data lakes, highly scalable orchestrated test execution and debugging, or convenient KPI visualization and analysis of test results in the cloud. In addition to the integration of custom software replay platforms, dSPACE offers its own platform solutions for the testing of software components and virtual ECUs for ADAS/AD applications at all development stages. For example, RTMaps, an easy-to-use multisensor development tool with integrated synchronous data communication, is an excellent performance-optimized solution for fast replay. RTMaps offers extensive libraries with function blocks for reading, recording, and preprocessing sensor data as well as for the simple integration of algorithms from Python, C++, Simulink®, ROS, and components according to the FMU standard. For even more complex replay, i.e., when a test candidate is available in the standardized form of an AUTOSAR Classic or Adaptive V-ECU and manipulation or filtering of data is required, dSPACE offers the right solution with a replay platform based on the offline simulator VEOS. V-ECU or FMU-based replay with VEOS supports CAN/CAN FD/Ethernet bus simulation and communication and allows an easy transition to closed-loop SIL for scenario-based testing, or in a later development stage, also to hardware-based testing using HIL.