Driving Simulator for Connected Autonomous Vehicles

According to the World Health Organisation, the number of traffic accident-related deaths remains unacceptably high, with over 1.3 million people dying each year globally. This number continues to rise due to growing population and increasing motor vehicle uptake. It has been reported that vast majority of trauma-related fatalities occur in the prehospital phase, while nearly a half of those are preventable.

Therefore, timely and precise crash detection systems and appropriate response from emergency services are critical for reducing preventable deaths. Existing systems are imprecise, unreliable, and computationally heavy.

An informal ongoing collaboration with Volkswagen, established during the summer project, allowed to explore multiple computer vision methods and deep learning algorithms for accident detection applied to an existing traffic surveillance infrastructure. The project team has been authorised to access every traffic surveillance camera in the city of Carmel, Indiana.

This allows testing deep learning algorithms on real-time data around-the-clock, during different times of the day and weather conditions, and on varying road layouts. As an outcome of the summer project, fundamental principles of computer vision and deep learning were explored and multitude of software tools, necessary for the upcoming PhD research, were utilised.

The groundwork for my own algorithmic development for this ongoing collaboration was established through deployment and benchmarking of state-of-the-art algorithms.

With human error being the predominant cause of vast majority of road accidents, computer vision-based accident detection systems can reduce preventable deaths. Once the appropriate level of technological sophistication is reached, connected autonomous vehicles will eliminate the human factor. The major PhD research is focused on contributing towards developing an autonomous driving simulator, which is an effective tool for generating and validating advanced driver-assistance systems.

Ongoing developments with Volkswagen provide technical training and research data that will be used for the simulator. For instance, inference data from Carmel, saved in CSV format, will be combined with OpenStreetMap data and used for developing virtual and mixed reality scenarios and modelling road networks in OpenDRIVE format. Furthermore, computer vision algorithms will be modified and optimised for the vehicle camera.

The simulator will then be connected to the powertrain dyno to allow hardware-in-the-loop control of a physical vehicle and its subsystems.

For Volkswagen, this research aims to develop a traffic accident detection feature that would reduce trauma-related fatalities that occur in the prehospital phase following a traffic accident. For IAAPS, it will contribute towards putting together a hardware-agnostic Advanced Driver Assistance System simulation software system with the ability to easily upgrade the hardware setup to support motion platforms and 360-degree projection screens.

With simulation testing being an efficient way of validating autonomous technology, such expertise will be beneficial for the institute, empowering the organisation to become a trailblazer in the automotive industry, making its inevitable future closer.