After we’ve heard the billions of dollars being pumped in to develop autonomous technologies, I’m sure there’s one thing we can agree – autonomous and affordable don’t go together. But there’s one Indian start-up that’s out to change just that.
Quite early on, Bengaluru-based Minus Zero realised two things – autonomous technology isn’t suited for Indian road conditions, and moreover, this technology is anything but affordable.
“The basic thumb-rule of the AI industry is that the more data you get your hands on, the more accurate you can train your model to be. This approach works very well when in case of say Google, that wants to show you relevant ads, since there is nothing at stake. But when it comes to a problem like self-driving cars, aka the Apollo mission of the AI industry, this data dependency becomes more of a curse than a boon,” says Co-Founder, CEO & CTO, Gagandeep Reehal.
“For starters, the companies and consortiums developing this technology would need to collect way more data than they’ve collected in the past 10 years. And to put this in perspective, in the last decade, they’ve collected millions of miles of driving data to train their model on and they are still stuck at let’s say autonomous Level 4. What does that mean on the road – it means that their system fails in unseen traffic scenarios. Let’s say if you bring one of their cars to India. There are high chances that it will fail in Indian conditions, because it has not trained on Indian data.
“This unpredictability makes it a very highly capital-intensive space. Large corporations have spent billions of dollars on this without actually getting a product out in the market. On the other hand, let’s compare this with humans. You don’t need to show a human baby 20,000 images of a bird for them to know that there is a bird in front of them. Humans have that inherent intuition by which we can extract maximum insights from a minimum amount of data. That was the intuition that we started mimicking.”
What started as an idea for a research paper a year and a half back paved the way for Minus Zero as it is today, thanks to the team’s conviction of bringing their solution to life.
Their journey began in May 2020, and since then it’s picked up funding from angel investors, gone on to receive a grant from the Government of India (GoI), it’s been a part of some accelerator programmes and its also been a part of NVIDIA’s Inception incubator programme.
“This industry has spent billions of dollars without actually getting a product out, so the general perception out there that people have is that this is a cash burning industry. There has to be some inherent problems in their approach that is making it techno-commercially not very feasible,” says Reehal.
“Leaving India aside momentarily, in emerging markets like Israel or Singapore, where 70% of the mobility ecosystem is concentrated, the solution has to be not just robust and one that considers the differing traffic scenarios, but it also has to be cost-effective. A single vehicle can cost anywhere between ₹2-2.5 crores (US$ 272k-340k) per vehicle. They’ve spent millions in R&D and their capital expenditure is very high. This is in no way to say that their approach is wrong, but it definitely isn’t feasible commercially.”
Going LiDAR-Free
Needless to say that Indian road conditions are quite different from the West. Even the most planned roads are fraught with potholes, making the existing road infrastructure anything but dependable. Then comes the willingness of people to follow traffic rules, which is barely existent. And when it comes to traffic deadlocks, that’s a whole different story altogether. So, you see, India is the toughest testing ground for autonomous vehicles.
“We’ve created our own proprietary AI stack, nature-inspired AI, that mimics human intuition. It’s the kind of AI that can extract maximum insights from minimal amounts of data, just the way we humans do. We’ve already filed a patent for this. Most autonomous companies use very expensive LiDAR 3D sensors to get 3D data of the surroundings – speed, acceleration and localisation of all the surrounding vehicles. Now, each of these LiDAR sensors cost around ₹9 lakhs (US$ 12,245.44).
“We, on the other hand, have eliminated the use of LiDAR entirely. Our proprietary stack can extract maximum insight from a 2D camera feed. For instance, I can get to know the exact speed of a vehicle in front or in the surroundings, its size, acceleration, predicted trajectories and more, all from a single monocular camera,” explains Reehal, “This is just the tip of the iceberg.”
“Most companies use an end-to-end neural-network based approach, where they consider driving as a single task and train on it. We, on the other hand, do it the way Tesla does and divide the driving task into a number of sub-tasks. These sub-tasks are powered by different algorithms based on the operational design domain, the environment in which the car will work and the kind of scenarios it will tackle. And so, it’s a combination of different technologies and 40-50 algorithms that together make the driving task possible in all types of scenarios.”
Eliminating Error From The Equation
A firm believer of collaboration between all players in this space, Reehal admits that India is still in a nascent stage in its mobility journey. With the government fully focused on realising the country’s electrification plans, it’s going to be a while before autonomous technologies get the attention that it deserves from the country’s lawmakers.
But he, like other Indian players in this space, is keen that the government start thinking about regulations in this space and make the environment conducive for companies to start actively venturing into this space.
After all, India is a country where more people lost their lives to road accidents than the Covid-19 pandemic. And with a technology like autonomous mobility, which eliminates human negligence and error from the equation, many lives would be saved indeed.
“You can expect our driverless vehicles to hit Singapore’s streets in 2024.”
As for now, Minus Zero has its hands full, with quite a few projects to look forward to.
“We completed a successful trial with a three-wheeler eRickshaw, which was a low-cost, makeshift solution to prove that our technology actually works. We’re now totally focusing solely on four-wheelers. Moving forward, we have two business models planned. First is the setting up of fleets of 15-20 driverless golf carts across private campuses like airports and educational campuses. The second model is a Robo-taxi or driverless taxis for public roads, which we’re in the process of building,” he explains.
“Our plan for campus mobility is to start beta-testing and providing early experiences to consumers by August or September next year. There aren’t any legal hindrances in private campuses, so that’s a plus. We hope to deploy those fleets by early 2023.”
“As for Robo-taxis, we will be ready for the market by 2023. But we’re hoping to deploy that project in Singapore, which means that we’ll take anywhere between 6 months to a year to clear the regulations. Taking that into account, you can expect our driverless vehicles to hit Singapore’s streets in 2024,” concludes Reehal.
