“We are planning to launch a product, named AutonomousOne, targeted towards December 2022 as a launch date. The autonomous driving version will be launched in mid-2023” 

DARPA (the R&D agency of the US Department of Defense) autonomous driving challenge videos by team MIT inspired Sanjeev, then a second-year undergrad at IIT Roorkee. The incident shifted his concentration towards academic research specifically focused on Reinforcement learning and Motion Planning till the time he completed his MS in CS from the University of Alberta in 2014.

The goal was to utilise the algorithms and mathematical formulations he has been developing to enable autonomous driving in the worlds’ most difficult traffic and environmental scenarios, i.e., the Indian traffic and Indian environments. 

“In 2014 I came to India deferring the Computer Science PhD offer at the University of Massachusetts Amherst in the US to 2015. When I came back, I finally decided it was time to actually start the company and in May 2015 I eventually registered the company and declined the PhD offer altogether. That’s how Swaayatt Robots started,” said Sanjeev Sharma, founder & CEO, Swaayatt Robots. 

Edited excerpts: 

Q. How you are leveraging AI/ML at Swaayatt Robots? 

Sanjeev: We have been researching in several different fields of theoretical computer science and applied mathematics and AI/ML happens to be one of the most prominent fields in which we are researching. Existing algorithms rarely work in practice, and we often have to invent new formulations or algorithms from scratch. For example, when it comes to deep neural networks, existing algorithms that achieve high accuracy are often cumbersome from a computational standpoint.  

This is where I started fundamental research in deep learning (again using my experience and mathematical knowledge base I had since my undergrad) and focused on developing novel loss functions or novel architectures that are significantly computationally efficient while achieving the same or higher accuracy compared to the state-of-the-art in the open scientific literature. 

Coming to the company at large, we are developing novel algorithmic formulations that can actually enable AVs to environments that are as unstructured to have the decision-making capabilities to negotiate highly stochastic-complex-adversarial traffic dynamics, such as in India. 

Q. What all solutions do Swaayatt Robots has in its bucket? 

Sanjeev: There’s still a lot to be developed before we claim level-4 or level-5 in India, but since 2016 I have been personally conceptualizing many ideas, centred around reinforcement learning (RL) and inverse-RL, to solve many of the complex traffic negotiation problems. 

• Developed the technology, when it comes to perception or algorithmic formulations (specifically focusing on deep learning) that enable autonomous vehicles to perceive their environment that is as unstructured as the environments in India, both during the day and at night, and that too without using LiDARs and RADARs.  
• We have developed a proof of concept of the technology that enables AVs to navigate without using high-definition maps, using just GPS maps just like regular human beings do.  
• Moreover, planning algorithms research constitutes 70% of the research at Swaayatt Robots, perception research another 25% and the remaining effort is in localisation.  

Q. How important is computer vision technology for the successful deployment of AVs? You have also launched a course in Computer vision via the Deep Eigen platform – another online education startup from you. Can you share the information and why CV holds is so relevant? 

Sanjeev: Computer vision is one of the foundational areas and having a good grasp of computer vision theory is fundamental to any company or team working on developing robotic or autonomous driving or ADAS perception algorithms. CV theory is very important if you want to be a researcher in such areas, or if you want to develop algorithms to enable autonomous vehicles to perceive their environments.  

Deep Eigen provides very deep courses as a structured graduate program course at the top 10 universities in the world. Typically, one would have access to such depth of knowledge only while pursuing an MS or PhD program in developed countries, typically in North American or European universities. In India, if one typically wants to gain advanced knowledge in the fields at large, there are no graduate programs that offer that, and one rarely has any option online. 

Our course - CV-1.0X provides very foundational knowledge to the audience at large so that they can excel in their careers in the industries or use the theoretical knowledge gained and algorithms learned through the course to build their academic research or industrial research career. Or they can just use the knowledge in solving real-world computer vision problems in the industry at large. One can find more info about the course here. 

Q. Can you share some of the difficulties you face when it comes to the field of autonomous driving? 

Sanjeev: Autonomous driving is going to be an R&D problem even 8-9 years from today. We have identified some of the very challenging traffic and environmental scenarios, which when addressed, solve the Level-5 problem.  

• One can develop highly accurate deep learning-based perception modules but the major concern is, what is the computational requirement of such algorithms. We are resolving and addressing those challenges, and coming up with novel deep neural network formulations and novel loss functions or compute units to be able to develop software that has a low computational requirement and also consumes less training data. 
• From day one, one of our goals is to solve autonomous driving in the world's most difficult traffic conditions, which happens to be in India. Once you ensure safety in such conditions, once your software has matured to that extent, you can ensure safety anywhere else in the world and don’t have to worry about corner cases in such structured traffic conditions. 

Again, hiring the right talent is one of the big challenges. We are also looking to raise the next round of funding to complete our seed-round funding requirements to grow at a much larger pace.  

Q. How well Indian market is placed when it comes to the autonomous driving domain? 

Sanjeev: Definitely there is a market for autonomous driving in India and it will start with the Advanced Driver Assistance System (ADAS). So, for the ADAS market in India, it is going to be at least a 30–40-billion-dollar market by the year 2025. This is where we are planning to launch a product, named AutonomousOne, targeted towards December 2022 as a launch date.  

The autonomous driving version will be launched in mid-2023. AutonomousOne will have level-3 autonomous driving capabilities and will also have the ADAS capabilities according to the European Union guideline for ADAS. So, we are following Euro-NCAP guidelines for developing the ADAS system.  

Secondly, the Indian trucking business will mature by the year 2035 and is going to be a 600-billion-dollar addressable market. So, the question is, how prepared are you when that event happens. Also, one of the entry points for autonomous driving is definitely going to be the defence. But if you are talking about the commercial passenger domain for Autonomous Driving, we are talking at least 10 years from today. 

Q. Any plans for the future? 

Sanjeev: Hereafter we would like to raise another round of funding as we have still not completed the target for our seed round funding. So, raising funds is one of the most important goals, it’s in the future plans. For now, we are also preparing for a 100 Km/h autonomous driving demo. Growing our team is also one of our important objectives and soon enough we hope that in a year we will be a team of 25. 

Image Credit: IT Web