Automotive Software Validation with Shift-Left and AI-driven Approach

Anup Sable CTO

The complex software and architectural requirements of software-defined vehicles are significantly impacting the software validation process. Anup Sable, CTO of KPIT Technologies, presents the challenges faced by the traditional V-cycle and introduces modern approaches and solutions such as virtualization and AI-based testing to overcome them.

by Anup Sable, CTO of KPIT Technologies
한글로보기







The complex software and architectural requirements for SDVs are impacting the automotive software validation process. The mature automotive software development process needs to be complemented with shift -left methodologies such as virtualization and enhanced by AI and gen AI driven testing. The focus of this article is toi elaborate firstly on the challenges that traditional V cycle poses and the new-age solutions to resolve these challenges. With this, we aim to help OEMs to achieve their SOP targets by avoid delays related to software validation.

So, what are the factors that lead to the increased complexity in SDV development and validation process?
What makes software validation difficult is the increase in software applications with complex and dynamic deployment. Validation of future software updates only adds to the burden. What are the factors that increase this complexity?

Firstly, multiple heterogeneous software and hardware components being present in vehicles leads to more than three times the overall test scenarios. Multiple domains including AD/ADAS, Infotainment, Connected Solutions with high software complexity are integrated into a single High-performance chipset (HPC) which in turn consists of different middleware, operating system, and network components. Each of these also needs independent validation considering architectural constraints, compatibility, and integration. 

Secondly, analysis & isolation of issues is tougher and even slower due to increased interaction & integration points within the diverse components. The North to South (N2S) and South to South (S2S) communication between the HPC and the multiple zonal ECUs (electronic control units) comes with unprecedented architectural/ network constraints and scenarios. 
Thirdly, the test strategy has been tightly defined around Real ECU. The test setup being available only very near-to-finalized architecture for SDVs is too late, considering the increased interaction among numerous heterogeneous components. Additionally, contrary to the software development for traditional architecture, OEMs (Original Equipment Manufacturers) are approaching supplier contracts for hardware with a significant amount of uncertainty.


 


How does the traditional validation approach affect the new software-driven programs at OEMs?
The OEMs are facing several challenges leading to SOP timeline delays of more than 2 years due to late identification of software errors in the v cycle

1. Feature/function compatibility & integration issues come to light just a few months before the SOP. Ideally, this should be identified from the start of the development phase and should be known well before the SOP timeline. 
2. The dependence of the development process on real ECU (electronic control units) needs to be reduced, and it needs to be decoupled from supplier and real ECU.
3. It takes more than four to six weeks to complete system V cycle and report the issue back to the development team. This needs to be closed in a matter of days.
4. The isolation of issues and identification of the root cause is difficult and takes more than a month  s time.
5. Insufficient test coverage including corner cases as one transitions from conventional programs to SDV programs


 


How can we complement the existing Validation Process with some shift left and AI-driven solutions?
OEMs need to foster a comprehensive test strategy that covers all available technology options from Virtualization and Reference Hardware to Real ECU. Below is a potential list of components that need to be considered to build such a strategy 

1. Virtual validation: The existing test strategy is tightly linked with real ECUs. The focus is on stopping the issues from escaping to production. While this still needs to remain the focus, the SDV SOP timeline pressure has increased manyfold due to the reasons stated earlier for SDV. We must think beyond traditional physical validation environments if we must keep pace with the increased volume and complexity of test coverage. Hence, it is necessary to make available a reusable virtual platform and increase its use for test coverage across programs. The usage of cloud-native environments and virtual platforms to complement and left-shift validation across V cycle. Automotive OEMs are spending millions of dollars to create a virtual platform for development and validation. However, investing without a transformational change within their organizations will be a recipe for disaster.  Dedicated teams need to own and drive this new process within the existing V cycle rhythm. 

2. Early Architecture & Platform Validation: It is crucial to validate architecture definition, platform & domain hardware constraints at early stage of architecture definition. The architecture design concept should be clear from the start despite the agile approaches. The process and scope of integration are often massively underestimated. To prove the concepts, integration must be started as soon as possible with the right eco-system and not only after real hardware is available. Moreover, the communication system/network is complex and must be viewed in a differentiated manner since not all traffic (Ethernet <-> CAN) is the same. Also, the safety and security aspects must be considered from the beginning. The availability of a reference hardware setup/prototype with SDV architecture can help in validation of network specifications, architecture decisions, and assumptions during PoC (proof of concept) phase. This will help OEMs reduce risks by developing hardware prototypes during the architectural PoC (proof of concept) phase and advance the pre-SOP to SOP development cycle by 1 to 2 years.

3. Integrated multi-domain HILs (Hardware in Loop) systems with different manifestations (NW validation, Architecture validation., Integration validation) will be crucial to test interdependencies across various domain functions in the context of SDV. Individual domain HILs will not be sufficient to test the interdependent behavior of different domain functions. Integrated HIL (Hardware in Loop) replicates real-world scenarios by testing all components in an integrated environment and ensures seamless interoperability with real-world network conditions. It also helps accelerate development by enabling parallel testing in a centralized system. They also seamlessly accommodate and test new feature releases ensuring up-to-date system validation. These HILs are further enhanced by advanced analytics that predict system behavior and potential cascading failures.

4. AI-driven testing - With the recent advent of AI and gen AI, it is imperative to utilize these technologies for productivity improvements. A huge amount of engineering effort can be minimized with both automation and AI generated test cases. The AI solutions can be augmented to existing tools and frameworks to reduce costs significantly. 

5. Increased Test coverage to include corner cases - One needs to come up with a comprehensive set of test scenarios considering the software explosion that haven  t been considered before. This will mean creating a net new set of test scenarios which need to be anticipated based on the new architecture and through learnings from diverse SDV programs.


 


How is KPIT helping its clients address the challenges in SDV Validation?
KPIT innovations will help OEMs make their validation strategy SDV ready by creating an integrated test coverage across E/E architecture, vehicle network, platform, multiple domains, and the vehicle. KPIT provides the complete spectrum of platforms and accelerators in the SDV validation process including the components mentioned earlier in virtual engineering, AI and gen AI, multi-domain system validation, Reference SDV architecture and network testing in the pre-SOP phase. Moreover, we help OEMs focus on seamless orchestration of tools, infrastructure (physical and virtual), formulating an integrated Strategy from CI/CD to Virtual to Physical Validation and creating an early roadmap for program-level implementation.

---