Ola Electric autonomous systems engineer, Mankaran Singh, recently converted his modified-Maruti Alto K10 into an autonomous vehicle using a second-hand Redmi Note 9 Pro. This was made possible with the help of Flowpilot, an open-source driver assistance system created by him. 

Flowpilot is an open-source fork of Comma.ai’s OpenPilot that can run on most Windows/Linux and Android-powered machines. In an exclusive interview with AIM, Singh said that he didn’t train driving models on his own; instead, he used learning models from Comma.ai because one needs tons of data to train models and require millions of dollars for compute clusters to train them. 

The idea to start this initiative stems from George Hotz, the founder of Comma.ai, which is also into enabling autonomous vehicles with the help of smartphones and other proprietary devices. 

An avid programmer and autonomous system enthusiast, Singh is also into gaming and developing web applications. Besides Ola – which he joined three months ago – he is also working on his passion project Flow Drive, the think-tank behind Flowpilot. 

Flowpilot performs the functions of Adaptive Cruise Control (ACC), Automated Lane Centering (ALC), Forward Collision Warning (FCW), Lane Departure Warning (LDW), and Driver Monitoring (DM) for a growing variety of supported car makes, models, and model years maintained by the community. 

It records data from road-facing cameras, CAN, GPS, IMU, magnetometer, thermal sensors, crashes, and operating system logs. “All you need is basically actuators, the controller steering and gas brakes to control the car. And if you have that, Flowpilot can essentially run on anything,” said Singh.

“It supports all the phones (Android) that have OpenCL supporting them. That’s basically the GPU drivers, which are used for image processing, like neural networks, exhibiting neural networks.” he added. 

Singh said that the quality of the experience definitely depends on the phone you use. “Mostly, if you’re using a phone that costs around 20-25K, that’s enough to run Flowpilot for reasonable performance, but anything less powerful, it’s going to lag, and the system will just show a warning; it just won’t engage,” he explained. 

Flowpilot supports over 200 cars, including brands such as Honda, Toyota, Hyundai, Nissan, Kia, Chrysler, Lexus, Acura, Audi, VW, and more. Even if a car is not supported but has adaptive cruise control and lane-keeping assist, it’s likely able to run Flowpilot.

However, Singh said that it does not support cars as of now in India. “In India, none of the cars are supported in a plug-and-play fashion. The only reason people use Flowpilot is because it’s far better than the stock system of the car.” he said

How it Differs from Tesla

“Tesla has a long term vision. They are using eight cameras and complex neural networks, including advanced 3D perception. However, dealing with eight cameras brings a much bigger computational load. While a smartphone can handle images from two or three cameras well, it gets challenging with eight cameras, all running at a fast 50 hertz. This requires real-time processing on a special chip designed for cars,” explained Singh. . 

He believes that Flowpilot has great potential on highways. “Highway driving can be automated with as little as one camera or a forward-facing camera and a bunch of other sensors, maybe GPS, IMU,” he said. 

Singh said that the immediate use case involves highway driving automation, which is advancing rapidly with features like ADAS (Advanced Driver Assistance Systems) becoming more prevalent in cars. “Many car companies are incorporating highway automation, where highways provide a more structured environment. In this scenario, drivers can sit back, relax, and let the car cover the entire highway.” he explained.

Other safety features include automatic emergency braking and blind spot monitoring. If a driver attempts to change lanes when an obstacle is present, the system will issue a warning. The machine learning-based braking system operates at a much higher rate than human processing, allowing it to brake at the right time and potentially save lives. 

As technology continues to advance, autonomous vehicles may find applications in private spaces such as college campuses and tech parks. These environments, with controlled and predictable conditions, make them more suitable for autonomous transportation, shares Singh. 

Limitations

Speaking on safety Singh said there will be people who would be retrofitting custom motors on their steering, which he does not recommend at all. “The only recommended approach, given that your car doesn’t support Openpilot or Flowpilot, is to just replace the steering with the one from a supported car. Because these actuators are safety-critical” he explained

Though Flowpilot is a good starting point for autonomous vehicles in India, it has several limitations as well considering the condition of Indian roads. As Flowpilot relies on mobile cameras, unlike higher-end systems employing LiDAR and additional cameras, it struggles with low-light conditions, shadows, and unpredictable objects like cyclists or animals. This vulnerability in perception can lead to misinterpretations of the environment, potentially leading to accidents.

Moreover, India’s road infrastructure varies greatly, with many roads lacking lanes, signage, and proper markings, posing challenges for autonomous vehicles to navigate accurately. Currently, Flowpilot can prove highly beneficial on highways, where traffic is comparatively lighter than in cities, and designated lanes for vehicles are available. 

What’s Next?

Singh said that he doesn’t have any interest in commercialising Flow Drive. “It’s not a commercial company. My team and I don’t have any interest in commercialising it. It’s open-source software, given back to the community,” he said, adding that next he plans to open something in the robotics space, focusing on automation, manufacturing, or warehousing because a lot of the underlying tech is the same.

Regarding Flow Drive, he said it will remain alive. “We will be supporting people, helping the community build more stuff on top of it, but it runs on donations, and that’s how it will always be,” he concluded.

The post [Exclusive] Ola Employee Makes Alto Autonomous with Redmi appeared first on Analytics India Magazine.