• Single system for accurate replay of sensor data into multiple SUTs in parallel
• Reduced complexity and maintenance of the overall test setup
• Data replay with lower reduced costs and space requirements without compromising quality
• Easy shift from less complex to more complex perception ECUs without the need to change test equipment 

Task

Ensuring safety and the reliability of vehicle sensing under all driving conditions is a crucial task for ADAS/AD systems at higher levels of autonomy. Accurate perception is a mandatory prerequisite for all subsequent steps in the data processing chain of autonomous vehicles (AV), from data fusion and localization to path planning and vehicle control. Regardless of whether it involves a small smart sensor like radar or lidar, a domain controller, or a high-performance computer (HPC), perception needs to be validated on target hardware using real-world data to ensure proper performance and robustness in different traffic and environment scenarios.   
 

Challenge 

Validation of a smart sensor or a small perception ECU with just one camera and a few CAN channels requires a less complex hardware-in-the-loop (HIL) replay system than for an ADAS/AD domain controller with a higher number of cameras and Ethernet channels for the provision of radar and lidar sensor data. One of key challenges for the customer is to choose the right testing environment. One possible approach is to start with an entry-level replay system to address low requirements of replay into smart sensors, then spend additional money and time to establish a more comprehensive replay solution to validate ADAS/AD controllers or HPCs in later projects. A different strategy could be to make an investment just once in a flexible hardware reprocessing system which can be used for testing a wide range of complexity of perception units in a variety of projects to secure the investment over a long time. 
 

Solution

dSPACE offers a modular data replay solution as a combination of scalable components consisting of a replay PC, raw sensor data, and bus data injection units providing a variable number of replay channels for flexible use. 
A cost-effective option with a focus on maximum efficiency is data replay into several systems under test (SUT) at the same time using a single hardware reprocessing system. The validated perception units can be of the same type, or they can differ regarding the quantity of raw data interfaces such as camera and bus channels for injection of a lidar point cloud, radar object lists, and further sensor payloads. The data injection into each individual ECU is done via separate channels with independent time synchronization. In the first step, the recorded and annotated sensor data is usually downloaded to local SSDs in advance, or it can be directly streamed to the replay PC from the cloud. RTMaps, the dSPACE high-performance execution framework for multisensor applications, runs on the replay PC as replay software to read out the recorded data streams within a wide range of file formats, such as MDF4, rosbags, MCAP, or PCAP, and to push the data into the data injection interfaces, the Environment Sensor Interface (ESI) unit and SCALEXIO. To ensure independent replay, a separate replay entity of RTMaps is executed as a process or in a Docker container for each device under test (DUT). dSPACE also provides services for converting data from one format to another that fits the specific SUT and for lossless and lossy data decompression, e.g., using H.265 on GPU. RTMaps can also be utilized to calculate key performance indicators (KPIs) after the replay by comparing the output of the SUTs with reference data. The entire data replay system can be remotely operated and monitored using gRPC and XIL API.
 

A single FPGA-based Environment Sensor Interface Unit (ESI Unit) offers a capacity for the injection of raw sensor data into up to 16 camera interfaces using MFC modules of all relevant types, e.g., FPD-Link, GMSL, ASA Motion Link, or MIPI CSI. The modules can be operated independently, which allows the ESI unit to replay raw data streams in several SUTs in parallel. In contrast, the SCALEXIO LabBox real-time system, or alternatively the SCALEXIO Processing Unit if more processing power is required, maintains a low-jitter real-time replay of frame-based bus sensor data, e.g., a lidar point cloud, and establishes the required restbus simulation to provide vehicle status info and keep the DUT at the expected operating point during the entire replay. SCALEXIO supports standard and automotive Ethernet (UDP, TCP/IP, SOME/IP), CAN/CAN FD bus networks, LIN, and even FlexRay. Similarly to the ESI unit, each bus channel in SCALEXIO can be operated independently, which allows the injection of bus data into several DUTs at the same time with a single SCALEXIO unit. The SCALEXIO LabBox or SCALEXIO Processing Unit also keeps the data replay accurate thanks to HW-based gPTP synchronization of all data streams, which is an essential precondition for successful validation of perception. The SCALEXIO unit typically acts as the PTP timeTransmitter and establishes separate synchronization of replay to single SUTs based on its real-time clock. The ESI unit and potentially also SUTs are PTP timeReceivers. However, this also works in a different way with a DUT acting as the PTP timeTransmitter. A combination of a single and modular ESI unit and a SCALEXIO system synchronized by gPTP for replay in multiple SUTs offers high efficiency and an optimized cost-performance ratio. Without compromising quality, it is also less complex, consumes less space, and requires less maintenance effort in comparison to the use of several low-level replay systems in parallel, each injecting data into just one SUT.

A further advantage is flexibility when shifting from one project to another with the need to validate more complex perception ECUs like an ADAS/AD domain controller or HPC. In such cases, the proven testing equipment can remain unchanged, since it already provides sufficient performance to establish a high number of data streams for raw and bus data injection. Even if the demand increases further in the future, the setup can easily be extended by further ESI and SCALEXIO components to meet this demand thanks to its high scalability. Another benefit of using dSPACE hardware reprocessing solutions is that the test system can be adapted with minimal effort to support closed-loop simulation and scenario-based testing.