Accelerating autonomous driving with virtual validation

Current scenario 

Real-time vehicle testing, a big challenge for industry.

With increasing fatal road mishaps, traffic concerns, and fuel emission issues, manufacturers are looking to build secure, sustainable, and personalized vehicles with a focus on machine precision over human competence. However, the scope for disaster in autonomous driving is yet to be marginalized.

To ensure enhanced accuracy in obstacle detection, environmental awareness, data analysis, and decision-making for vehicle control, autonomous vehicles will need to be driven hundreds of miles before we accept their performance and reliability. This is an iterative process with a high dependency on manpower for real-time testing. In addition, the vehicles need to be tested in a wide range of scenarios before deployment. This could mean long waiting periods given that certain scenarios such as a snowy road, hailstorms, heavy rains, and traffic congestion are difficult to cover in real vehicle testing. Real-time testing in such rare scenarios is time-consuming and capital-intensive. Undoubtedly, a bulk of auto players' resources goes toward software R&D. This is where virtual validation, or the testing process in which the environmental test conditions are virtually simulated, comes in.

Virtual validation

Making testing easier and affordable.

Automated vehicles must be tested in every possible complex scenario. Testing all systems in the virtual environment rather than validating all the scenarios in the physical environment using actual vehicles will be cost effective and time efficient. However, manually creating virtual test scenarios would require domain knowledge of all possible situations. An easier way out is to shift to virtual validation by performing digital validation of AD systems. To perform virtual validation, we need to create virtual test scenarios that cover all operational design domain (ODD) parameters like weather, traffic signs, and road geometry by incorporating the National Highway Traffic Safety Administration (NHTSA) framework, New Car Assessment Program (NCAP) test protocols, and International Organization for Standardization (ISO) standards. This creates a database of environments in which a vehicle will be driven in real-time test scenarios.

Original equipment manufacturers (OEMs) collect a huge amount of sensor and vehicle data from test drives. This requires them to perform extensive physical testing to test the reliability of the autonomous vehicle in different environments. And this is hugely expensive. In such a situation, virtual scenarios can be automatically created from vehicle recorded logs using AI, where software can be first tested in a virtual environment and then fine-tuned further, reducing the dependency on test drives and the workforce.

Validating efficiently

Rain or shine, simulation-based validation can create different scenarios for testing.

Different environmental and traffic conditions can also be created to generate multiple combinatorial test scenarios for ODD coverage. For example, vehicle data recorded during clear sunny weather on a less crowded highway can be transformed into a digital scenario with dark cloudy weather along with traffic jams. Such scenarios, if generated in open standards, can be further imported with third-party simulation tools like NVIDIA DRIVE Sim, IPG carmaker, dSPACE ASM, PreScan, and Carla. This digitally transformed data can be reused to validate future software releases with minimal real-time vehicle testing. This will further enable AD systems to be validated robustly and efficiently.

Toward a safe future

Reducing human and machine errors with highly accurate data handling platforms, advanced precision algorithms, and AI techniques.

To get to complete autonomy, auto players need highly accurate and efficient platforms for data handling. They should be able to implement advanced precision algorithms and state-of-the-art AI techniques for better safety and travel experience. This can effectively reduce human or machine errors. Virtual testing in AD can help automotive players to accelerate the process of testing for performance and safety and help auto-makers achieve early time-to-market while reducing vehicle test drives, resource dependency, and costs.

To accelerate validation of autonomous vehicles, we need digital transformation by creating virtual test scenarios ensuring the coverage of all Operational Design Domain (ODD) parameters.