Slow to start,
Indian tech startups,
automakers in full throttle
on autonomous vehicles

Hi-Tech Robotic, which is working with at least half a dozen automakers, offers a good example of what stage autonomous vehicle technologies is in India. “In 2015, carmakers became keen on commercialising autonomous technology. We are working with six to seven OEMs – like Tata Motors, Mahindra, Maruti Suzuki – for both passenger and commercial vehicles,” says Ritukar Vijay, head of autonomous technologies and business strategy, Hi-Tech Robotic.

Vijay said that one of the biggest reasons why deployments have been slow is cost. Going by current costs, he estimates, a fully autonomous vehicle will cost as much as 10 times of that of a regular vehicle making it an unviable product.

As more and more automakers have started working on autonomous vehicles the cost of technology sensors have dropped from $75,000 to $10,000 over the past six years. “Between 2022 and 2025 the cost of the laser sensors will be $400-500,” says Vijay.

He wants to trim costs even more. The future is in vision technology as against the extant radar technology, he says, without disclosing costs. “Earlier, the laser sensors provided a lot of data, which is not useful in commercialisation of the technology. It is good for development.”Experts also believe that lower costs will democratise technology. The situation was similar with automatic cars. Until Maruti launched the AMT gearbox, a new gear technology for automatic cars that was marginally more expensive, less than 1% of cars sold in India were automatic. Today, they make for 5% for car purchases here.

Some features like self-park assist are already available in luxury cars like the Jaguar and the Mercedes, where the car parks itself without any human intervention. “The concept is ready… it is 90% ready. It is very difficult to crash a car with those sensors,” says Deepesh Rathore, London-based analyst of Emerging Markets Automotive Advisors.Unique India’s complexities – ranging from poor marking and signs to traffic that rarely keeps to lanes to cattle on roads – makes autonomous vehicles a money-guzzler. “The Indian challenge is that where will the technology get its reference points from… roads do not have marking nor is there any [respect for] traffic rules,” says Rathore.

That’s where artificial intelligence (AI) and machine learning (ML) will play a role, making it an ideal playground for companies such as Bengaluru’s Tonbo Imaging to train its system on unstructured traffic situations. “A lot of companies are trying to train their systems in places like California, where they have good roads and structured environments. How will you take vehicles out of there into unstructured environments... like China and India," asks Ankit Kumar, Tonbo’s CTO and co-founder, who is among the few Indians to have taken part in the famous DARPA Urban Challenge of 2007, the first long distance competition for autonomous cars.

Hi-Tech Robotic’s Vijay says that vision sensors combined with AI will make autonomous tech really competitive. “Our core sensors are camera sensors, along with we apply computer vision techniques and ML techniques,” he says. The company’s solution: a camera identifies an object and the machine learning module helps recognise it – whether it is a bike, car, a human, or a cow.

If it is a person crossing the road, the solution will stop or steer accordingly. If it is another vehicle, the ML technology will not only recognise the vehicle but also calculate its speed and distance, and apply brakes or slow down.

For example, if it's a large truck, it will keep a wider distance between the two vehicles. In case of agriculture and for Mahindra, the technology will help identify if it's a stack of hay in the corner or scarecrows in the middle of the field.The technology can predict collisions based on driving patterns. In the west, accidents due to lane violations are rare but common here. “n India, we are trying to solve that problem with a couple of million images and videos. We are trying to learn from the data that has been collected over time,” Vijay said.

All autonomous vehicles rely on technology to be self aware of their surroundings and navigate. They usually rely on a radar for detection of objects, people and other traffic, a camera to read signs and traffic lights, and a GPS for basic navigation.Standard cameras can see visible light and colours but will not perform well in night time, rain, fog or snow.
Radar, short for radio detection and ranging, helps with detecting obstacles, their range, velocity and angle. They work in bad weather and poor visibility, too, as they use radio waves for detection. Lidar – short for light detection and ranging – goes a step forward and helps create a precise 3D map of the surroundings. As they use light for detection, Lidars tend to perform poorly as visibility and weather deteriorate.Some of these technologies are being used by startups to built subsystems and technology for autonomous vehicles. Tonbo Imaging, which develops sights and scopes for the defence, has also developed a variant of its imaging system for autonomous vehicles.

“Our thermal imaging system is not light-based. It operates in the 8-14 micron wavelength range – this means that it can penetrate through fog where a Lidar cannot see. But a Lidar also comes with its own advantages, so we have combined sensor package to take advantage of all systems,” says Arvind Lakshmikumar, CEO and co-founder, Tonbo Imaging, which raised $17 million funding in September.

Lidars are expensive and complex mechanisms. They have to rotate to get 360-degree information around the car. Even though today there are solid-state Lidars available in the market which have less moving parts and are more cost effective, companies such as smart electric carmaker Tesla have done away with Lidars and rely on a system consisting of just ultrasonic, radar, and visual imaging cameras for maintaining the current level of autonomy it provides.

Tonbo has developed a sensor package that consists of a visible sensor, a Lidar, and a thermal imager. The data collected by Tonbo's sensing system is analysed, classified and acted upon by the processing unit developed by the company.

But, the big problem is how to deal with a vehicle or object that the system cannot see or detect, says Lakshmikumar. To help with such situation the company has implemented a vehicle-to-everything (V2X) communication into its system. V2X communication works directly between a vehicle and other infrastructure, objects, people, and other vehicles in its range. The big challenge for the V2X system is that it will function well only with its wide adoption and standardisation.

Currently, Tonbo is using its system for its military applications like autonomous armoured vehicles and trucks. "Our imaging system for autonomous vehicles have been deployed and are being tested on armoured vehicles right now," says Lakshmikumar.

Tonbo counts Israeli defense manufacturer Rafael Advanced Defense Systems – different from the Dassault Rafael of France – and heavy vehicle maker Caterpillar among its customers. “Over the next year we intend to take this to automotive customers,” says the CEO. Tonbo's systems use Qualcomm chipsets for processing the data; Qualcomm is an investor.

Steradian Semiconductors, a Bengaluru fabless semiconductor company, has developed a high-resolution 28nm millimeter wave imaging radar system, which the company claims is the world's most compact system. “With some clever processing, whatever data is required for autonomous vehicles can be pulled from a radar and visual imaging system,” says co-founder Apu Sivadas. “Our system along with an imaging camera is sufficient for autonomous operations.” The system differentiates a truck from a hatchback by identifying the shape of the vehicles.

Radars can see through bad weather or visibility. Steradian’s system, which consists of a chip, antennas and a software algorithm that learns and processes data collected by the antennas, offers higher resolution and a better point cloud of objects as compared to traditional systems.

Sivadas says, “Our systems are designed in such a way that per IC we allow for a larger sensor count thereby having a bigger array of antennas which result in better imaging.” The IC designed by the company has a much smaller die making it more cost effective. The company claims that the system provides imaging up to 40 to 50 meters and can detect objects up to 300 meters.