Less is more, goes the popular adage. And this seems to be catching on in the field of AI as well. Over the years, machines will become less artificial and more intelligent, as they wean off big data models and resonate with top-down logical reasoning, essentially improving the scope for making AI horizontal and integrative.
There are many limitations surrounding data-rich business models – high compute power that is expensive and harmful to the environment, high capex costs involved in adding various kinds of hardware to collect data, the process of data collection itself which is cumbersome in countries that don’t have clear cut data protection and collection laws and so on. Even if companies can rally all these resources, there is still no 100% guarantee that AI will provide secure or usable solutions. Take the case of autonomous vehicles – driverless cars are trained on millions and billions of datasets. Waymo has collected 20 million miles of data, while Tesla has collected nearly three billion miles worth of data. In addition, Tesla has been utilizing a neural network training supercomputer called DoJo atleast since 2019 to process large amounts of driving data.
But these data-rich neural networks cannot always handle edge use cases, which require decisions to be taken on the fly with limited data. This can cause serious anomalies and even accidents – the fatal accident in Arizona caused by a self-driving Uber killed a passerby and cast a huge shadow on autonomous vehicle development everywhere.
But an Indian startup has seemingly solved this issue, claiming to be India’s first startup building a fully autonomous vehicle, powered by proprietary nature-inspired AI and not heavily dependent on huge datasets or expensive hardware like LiDAR. Bangalore- based Minus Zero, cofounded by Gagandeep Reehal and Gursimran Kalra, has already tested their theory on an e-rickshaw, which navigated its way on Jalandhar’s roads without a driver last year, on a shoestring budget of Rs. 50,000 and four months to prep.
When asked how he arrived at a low-data, limited hardware model to produce a fully autonomous vehicle, Reehal says the idea stemmed from a research paper he and Kalra worked on while in college. “Data dependency is the biggest bottleneck to the autonomous vehicle space and is still a long way from commercialization as a result. Unlike ad personalization where different kinds and large amounts of data makes the customer experience better, self-driving should be able to deliver accurate results even when there isn’t a lot of data to work with.”
Relying on inherent human intelligence that can apply skills across domains, Minus Zero is building capabilities that can enable vehicles to learn from natural surroundings and apply the basic physics of driving. This approach is similar to what Tesla is experimenting with – dividing the big, overarching task of driving into smaller tasks and perfecting these tasks with utmost precision and accuracy. Not using LiDAR is a bold move, but Reehal’s company has developed an algorithm that can extract 3D data using 2D cameras – they have already filed a patent for this application. The end-to-end platform offered by Minus Zero comes with eight cameras, five radars and 12 ultrasonics. While the eight 2D cameras function as the ‘eyes of the car’ and gather data, the radars and ultrasonics work as backup data collectors in case one or more cameras fail.
Additionally, the platform engages in sensor fusion, image calibration, obstacle detection and perception gathering, mapping & localization, motion and mission planning, route optimization and a feedback-based control actuation loop. With a hybrid neural network and considering cognitive biases to finetune predictions, Minus Zero can drive safely even in random traffic conditions.
Reehal says, “Indian traffic conditions are among the most complex to master, and a driver’s skill heavily relies on his ability to think on his feet while driving. Our technologies reflect that style of thinking and skill level.” By 2023, Reehal feels they can achieve Level 5 autonomy, which he concurs is an aspiration for the entire auto industry. In the next 10 months, the team is hoping to retrofit a third-party EV with their technology stack and bring up their tech to Level 4 – where players like Waymo and Cruise are currently operating at. The startup is hoping to work with third party OEMs who can build a car based on the tech specifications they can provide. This is a lucrative proposition for OEMs seeking to get into the AV space in India because Minus Zero offers nearly 10x reduced R&D costs, 5x lesser capex costs on a single vehicle compared to industry counterparts and need around 20,000 kms of data for their platform to work anywhere in India.
With the goal of shared mobility in electric vehicles, Minus Zero hopes to see their tech rolled out in office and university campuses very soon. Emerging markets like Indonesia, Israel and Singapore other than India is on Minus Zero’s radar, for these countries are leaders in ride sharing and fostering vehicle aggregators. “Ultimately, we want the larger public to ride in autonomous vehicles, like they do cabs,” explains Reehal.
Minus Zero is currently supported by Nvidia Inception, AWS Activate and MeitY’s TIDE Grant, and comprises of a team of 25 engineers.
Source: indiaai.gov.in
