Data replay, also called data reprocessing, data playback, re-simulation, or HIL open-loop (HOL) is a key methodology for testing and validation of perception and sensor fusion components, mainly for autonomous vehicles (AV) in the automotive area, but also in agriculture, off-highway, railways, and aerospace. Due to the usage of real data, data replay is on the same level of realism as real test drives.    

Challenges

Along with the increasing level of automation, the complexity of the software stack and hardware for autonomous vehicles increases exponentially and, according to expert opinions, will lead  to doubled V&V (verification and validation) costs within ten years, in comparison with current development. 
Sensing and planning technologies are especially seen as most critical and must be tested intensively. Perception, sensor fusion, and object detection algorithms for sensing are mainly based on computer vision and require large-scale testing using real-world sensor data to meet functional requirements and ensure the expected level of safety and reliability.  
 

More than Pure Functional Testing

In addition to pure functional testing, the behavior of the SW stack deployed on the target HW has to be validated as well under realistic data loads and timing conditions, to ensure the correctness of interfaces and communication between real sensors and ECUs and to guarantee the required target system performance, safety, and security on roads.
To reduce V&V costs and time to market, validation and homologation need to shift more and more from real roads to the lab. Test resources for virtual validation and homologation must be connected to global data lakes, operable by worldwide teams in 24/7 mode, and flexible enough to cover frequent updates on SW, sensors, and ECU hardware, and changing board nets and security conditions. Ideally, this should be done by widely reusing exiting real sensor data, without a need for additional large-scale logging companies.