Indian startups compete in race for affordable autonomous driving, ET Auto

[ad_1]


<p>Representative image.</p>
Representative image.

Bengaluru-based robotics engineer Mankaran Singh posted this on X late last year: “My second hand redmi note 9 pro running flowpilot is driving my alto k10. Can it get more desi than this?” The video shows the small hatch speeding on a highway in pouring rain. There’s no driver.

The video, which went viral, was a 49-second clip of FlowPilot in action, an early iteration of what’s apparently a plug-and-play assisted-driving system being tested in India. Apart from global attention, it also sparked debate among engineers and enthusiasts on the merits of frugal innovation and possible safety risks.

Singh and his engineering batchmates Gunwant Jain and Raghav Prabhakar have developed what appears to be a version of something that Tesla, Google and others are spending billions to create. FlowPilot is an autonomous driving assistance system that can be operated with a laptop or a mobile phone. About 2,000 volunteers are testing it across the world, according to Singh.

“We started this project in the third year of our college, and were inspired mostly by George Hotz, known for reverse engineering the PlayStation 3 and developing iOS jailbreaks, and the founder of Comma.ai, which began OpenPilot,” Singh said.

OpenPilot is an open-source advanced driver assistance system (ADAS). Available in costlier vehicles in India, ADAS enables–among other things–automated lane centering, adaptive cruise control, lane change assist and driver monitoring.

While ADAS is not necessarily equivalent to driverless operation, Hotz has said that Comma’s mission is to “solve self-driving cars while delivering shippable intermediaries” and it’s “looking increasingly like we will win alongside Tesla and Mobileye.”

To be sure, Tesla and its rivals operate in countries where compliance with traffic rules is better than in India.

Motoring author and historian Adil Jal Darukhanwala points out: “This technology, as a demonstrator in an open field, works brilliantly. In a field dotted with pedestrians, two-wheelers and little regard for road discipline, it is a recipe for disaster… Our roads are mayhem at their best.”

Also, roads minister Nitin Gadkari has said that driverless vehicles won’t see the light of day in India, citing job losses for drivers.

That hasn’t fazed the makers of FlowPilot, which doesn’t need proprietary hardware, as OpenPilot does.

“We made it compatible with Android phones, Linux systems, Windows PCs. Now all you need is a laptop or the phone in your pocket. You download some apps and it becomes a radar to help drive your car,” he said.

FlowPilot’s demos on X and Discord show cars staying in their lane in heavy traffic.

“We have around 2,000 users and people have driven tens of thousands of miles. We get the driving data back to further train our machine learning models,” Singh said.

But getting it to work in India will need changes to be made to cars, apart from the absence of regulations governing such vehicles.

“None of the cars here are supported where you can just plug and play a phone for ADAS,” Singh said. “You have to do heavy modifications to get this thing running. But in the US and Europe, almost 50-60% cars on the road would be supporting this out of the box. So for people living in those countries, it’s much easier.”

The Alto K10 was modified with the steering system from a supported car. Singh also installed Panda, a universal car interface built by Comma AI that allows devices to issue commands and instructions to cars while in motion. This is also where the critical safety code resides.

Phones are perfect for FlowPilot.

“It has a front camera for monitoring the driver, while the rear camera is used to scan the road ahead. It has a GPU, a very powerful CPU, digital signal processing units, everything is there,” Singh said.

Cars loaded with the system can theoretically achieve the third level of autonomy in driving assistance–or L3 autonomy–as it’s called, a notch above that offered by premium cars in India in their ADAS systems.

The jump from L2 to L3 automation is substantial. Such vehicles can make ‘informed’ decisions on their own, such as accelerating past a slow-moving vehicle, or navigating a traffic jam. But they still require human oversight.

FlowPilot has local rivals.

Bhopal-based Swaayatt Robots, founded by IIT Roorkee graduate Sanjeev Sharma, has been working on the technology since 2016. His algorithms are a step ahead of Waymo and Tesla but lack of funding and adequate regulations have kept the startup in stealth mode, according to Sharma.

“Presenting autonomous driving in complex, stochastic and adversarial traffic-dynamics, on the roads in India,” Sharma wrote in a post on X, attaching a video that shows a Swaayatt driverless car navigating Bhopal city traffic at night.

Sharma questioned the path adopted by OpenPilot and FlowPilot.

“FlowPilot, which is a fork of Comma.ai, does behaviour cloning. Anyone who knows any bit of autonomous driving knows behaviour cloning is a dead end,” he said. “In this approach, you show the vehicle every possible scenario, infinite scenario, and then teach recovery from every possible action.”

Swaayatt takes a more end-to-end approach to the emerging technology, developing algorithmic models to pursue level 5 autonomy—where no human attention is needed–keeping safety and operational cost efficiency in mind.

The Bhopal-based startup has a team of 15 full-time engineers and 11 data annotators, with Sharma acting as the chief research scientist. Swaayatt has a fleet of SUVs fitted with off-the-shelf lidar, cameras and sensors.

“We are using a Mahindra Bolero which is easy to retrofit. We have designed our own electro-mechanical system and mounted it in the vehicle for steering, brakes and accelerators. We are currently preparing a Mahindra Thar for the next demo,” Sharma said.

Sharma says Swaayatt is the first company to successfully showcase the ability to negotiate bi-directional traffic on a single-lane road, where the vehicle may even need to shift off the road, and drive through uneven terrain. Sharma says this remains uncharted territory for larger companies dealing with the tech.

Swaayatt raised $3 million seed funding in July 2021, along with a $1 million research grant from the Ministry of Electronics and Information Technology (MeitY). Sharma is now looking to raise more funds before attempting a 100 kmph test on Indian roads.

Darukhanwala said these technologies could find a market overseas.

“The technology will be developed here. India is a hotbed for developing these complex algorithms, but to put them in use, they will have a better success rate in mature markets, where rule of law and traffic discipline is above everything else,” he said.

Swaayatt’s Sharma takes permission from the local police station and conducts tests at night. And, he’s always in the vehicle.

“The last amendment to the Motor Vehicle Act happened in 1988. There is no mention of ADAS or autonomous driving there, so in case your autonomous vehicle hits someone, it will be considered that whoever was sitting inside was responsible. That’s why in all our tests, I’m the one in the driving seat and I’m ready to take that responsibility,” Sharma said.

  • Published On Feb 4, 2024 at 08:37 AM IST

Join the community of 2M+ industry professionals

Subscribe to our newsletter to get latest insights & analysis.

Download ETAuto App

  • Get Realtime updates
  • Save your favourite articles


Scan to download App


[ad_2]

Leave a Reply

Your email address will not be published. Required fields are marked *