Vinay M K, VP-Engineering, PathPartner was inspired by an experience that hit closer to home for him. Road accidents in India are because of human errors and the negligence of the road users. Drunken driving could be counted as one of the many reasons but drowsiness and distraction too are close second.
Vinay proposed to bring forth a driver monitoring system to the management at PathPartner and started on a research on the topic. He then hired a former colleague, who had a doctorate and thus began the start of a team for the venture. Currently, it is a team of 20 people, with customers ranging from Bosch India—who employ the driver monitoring system in the trucks, to Motherson Innovation, Intel India, etc.
Vinay explains that the team worked on a module comprising of both hardware and software that could be licensed to automotive customers.
Behind the scenes:
PathPartner has seemingly simplified the driver monitoring system. The software as mentioned above keeps looking at the driver and alerts him if he feels drowsy and distracted.
But behind the simplified system is a process that PathPartner had to build for seamless functioning of the model. As the underlying technology is artificial intelligence, the model for which is built basis previously collected data, the challenge was the latter—data collection.
The data for PathPartner was the video that had to be collected from the trucks. There was unfortunately no hardware equipment handy at the time, hence the team got down to building its own hardware proto box. It took eight months and a partnership with Uber, Ola and Lyft drivers in the bay area to collect the necessary data.
The drivers had to put the box on the windshield and start recording their video sequences. Meanwhile the team of annotators had to go through the stream and mark the events—driver’s drowsiness or distraction—manually. It went through two levels of quality check. Notably, PathPartner partnered with an NGO providing livelihood to the specially-abled people for the annotation job.
Hence, more than 150,000 video streams were collected and more than 3,00,000 face images of different types of people and facial landmark annotated.
“This forms the base of research and algorithm development. We also stuck a partnership with the department of AI in IISc. The students cooperated by working on the data and built models in conjunction with them. These were the experiments we did in the early days. And now we have built a team of data scientists and the whole process involves data collection, annotation, model curation and optimizing the model for embedded platforms and writing the overall application,” Vinay says.
The road isn’t smooth:
The major challenges one has to face in the deep tech ecosystem as per Vinay is the availability of talent. “Data scientists are most sought after,” he says.
“Second, it is expensive to build the hardware prototype in India. The prototypes are usually in small quantities therefore the component availability, semiconductor support and manufacturing are less or expensive comparatively. The prototype building cost is high in India. Another challenge is India Inc. are not big buyers of deep tech. Indian IT companies are not investing sufficiently in deep tech,” he says.
The best way to work around Vinay says would be if the government allow Indian IT companies to invest in deep etch startups and provide tax ops or reduce capital gains for the same as form of encouragement. “The large companies have the reserves and the reach, and can help Indian deep tech startups reach large MNCs outside India. But to do that they should have some investment or skin in the game and that should be recognized as valid R&D investments,” says Vinay.