By Dr Roshy John, Practice Head, Tata Consultancy Services
I started thinking about the development of driverless cars as a hobby back in 2010, after having a near death experience in a hired taxi because the driver was over tired. The concept of driverless cars were emerging back then with a few references such as Stanford Stanley which won the DARPA Grand Challenge.
My research and career mainly focuses on Robotics. Back then I had developed robots for different industries and had a handful of experience in developing mobile robots. We have this practice of using 3D simulations to simulate a robot concept before prototyping it. I planned to follow the same path because doing a driverless research on an actual vehicle would be so expensive, much more than I can ever afford.
The first challenge was selecting a suitable vehicle. I made a spread sheet with the required parameters based on the initial research. Automotive industry is booming in India and there are plenty of choices. I took test drive for most of the potential vehicles which could be used for my project, hoping someday I would be making a driverless car prototype.
I found Tata Nano to be the best choice because of multiple reasons – though it is a small car from outside, it is so spacious inside and there is enough room for computers, actuators and sensors inside the car. Tata Nano is the only rear engine vehicle in market and there is a storage space in the front. My initial decision was to place the pedal robots inside the engine bay, there. There is another important reason for choosing Tata Nano. The On-Board Diagnostics(OBD) data is the only source for the driverless algorithm computers to know the state of the engine. In Tata Nano, it is easily readable.
Prototyping a driverless car
I thought of using an industry standard simulation system for the prototyping of the driverless vehicle.
The major advantages of simulation based development are:
- Almost all physical parameters can be simulated
- Physical Design iterations can be reduced which leads to drastic reduction in development costs
- Driverless algorithms can be developed in parallel and need not wait for the hardware to get ready
- Test cases can be run continuously
- Almost all sensors can be recreated inside the simulation
- Total code reusability – Theoretically the driving algorithms cannot distinguish whether the data is coming from a real vehicle or a simulated one
- Supports CAD drawings from most of the industry standard CAD development platforms.
- Most urban and rural roads can be created and the driverless car can be “driven” in it.
All I had was a simulator license and a few workstations. I didn’t have enough money to fund a complete driverless vehicle research like a University or a large company!
But, I was blessed with a team of wonderful engineers who were ready to contribute their time for this research as they were too bored with the regular monotonous work. We planned to work during late nights and weekends.
We got the sensor data sheets from the sensor manufacturer’s websites and we re-created it inside the simulator, with the same data protocol and in the connectivity sense. A life size virtual Tata Nano was created inside the simulator referring the product brochures available on Tata Motors website.
The driving algorithms worked on a separate workstation and the simulation on a graphics intensive workstation. Both these workstations were connected together using a TCP/IP network. We tried so many hours of simulation, test case execution and failure modes.
We did simulations for over 8 months, by which I was over 70% sure that this concept is going to work. By that time my financial side was also getting better. I was inventing a lot of stuff and started receiving incentives and royalties for the patents filed. My then company gave incentives and perks apart from salary for the patents granted. The excitement was tremendous which made me think about continuing this research on a real Tata Nano with real sensors.
I bought a Tata Nano and totally reverse engineered it to understand how to hack into the vehicle’s electronics to connect sensors, actuators and engine scanners. The major difference with simulation is that, we considered the vehicle to be having automatic transmission, but back then Tata Nano had only manual transmission. The initial task we took up initially was to convert the manual transmission to automatic – which is now known as “Automated Manual Transmission”.
The concept of driverless cars is not yet legalized in India. Thus, all the road tests were done on secluded residential projects – thanks to the non-occupied villa projects where the builders construct well defined asphalt roads before constructing houses. We did the initial pilot testing on these roads where the car did limited self-driving. We continued this research over these years and made a driving robot modular which can be fitted on to any vehicle. I have also filed the patents protecting the concept.
These days, almost all the automotive manufacturers are continuing with their own research on autonomous vehicles. This could lead to the development of new standards and new insurance models. But surprisingly, I find teaching the old dog new tricks (an autonomous driving robot) more fun. Over and above, using simulation reduces the development investments to a bare minimum.