Potholes on Indian roads: Can AI help fix them?

NEW DELHI – In August, residents in Hyderabad’s Hafeezpet neighbourhood were so fed up with a pothole disregarded by the authorities that they decided it was time for some public nudging.

A man stood in the large, craggy pothole and struck a yoga pose, dubbing it “Pothole Asana”, and uploaded the photos on social media. “Why spend on a chiropractor when our roads offer free spinal adjustments with every ride?” he added sarcastically.

The quirky protest went viral, and in less than 24 hours, the jab achieved what complaints had not in 60 days – the pothole was fixed.

Potholes have long been a blight on Indian roads. Thousands of people are maimed or killed each year, especially during flooding rains or in the dark when it is difficult to spot them. In 2023, crashes caused by potholes claimed 2,161 lives, up from 1,856 the previous year.

Indians have tried various means to fix them, ranging from painstaking court petitions to ingenious protests. Now a clutch of Indian start-ups are attempting something new: using artificial intelligence (AI) to detect and flag potholes to the authorities for repair.

In a fourth-floor office in Delhi’s South Extension, a neighbourhood dominated by garment and jewellery stores, workers inspect videos on eight large television screens, cross-checking visuals flagged for potholes and other road-related problems.

This is the nerve centre of Nayan, a firm that specialises in AI-powered road and traffic monitoring for government agencies. The footage comes from its dashcams in vehicles plying different Indian cities.

It is not alone. At least five other firms are deploying AI to get India’s shoddy roads fixed.

The Pune-based RoadBounce detects road defects for the authorities using an AI-enabled app on smartphones placed in its vehicles to record vibrations and geotag potential problem spots.

In August, Indian electric scooter company Ather also launched its pothole alerts feature for its riders, thanks to its road-condition maps built using data generated from vibrations of hundreds of thousands of its scooters in cities such as Bengaluru, Pune, Delhi and Mumbai.

India has the world’s second-longest length of roads, stretching to more than 6.67 million km. Monitoring such a vast network for defects manually is practically impossible, leading to persistent challenges such as poor quality of roads and a slow response time in repairing defects, if at all.

These AI-enabled tools, which cut down time and expenses required for road inspections, while also prompting the local authorities to become more responsive and accountable in maintaining them, could potentially redefine India’s infrastructure governance.

Mobile CCTVs

Each day, Nayan monitors roads in around 15 Indian cities for various civic corporations, using some 500 proprietary dashcams and 2,000 smartphones with its app.

Fixed mostly in government vehicles such as public buses and police patrol cars, these AI-enabled devices scan the roads for potholes and a range of other problems, including traffic violations and civic issues like uncleared garbage mounds.

“We have trained the model with hundreds of thousands of images of what a pothole or other potential violations may look like,” said Nayan head of sales Umang Gupta.

Depending on the client’s needs, the model flags the problems, along with geotagged visuals. The AI model is not perfect, at least not yet; there are times when a shadow is reported as a pothole. In case of such doubts, the visuals pass through a more advanced second tier of AI checks and, if necessary, even a third tier of human checks.

RoadBounce’s smartphone app, on the other hand, uses vibrations to measure the country’s road smoothness based on specifications set by the Indian Road Congress. Each time a vehicle vibrates more than it should, that specific spot is flagged as a problem.

“Imagine a step counter for a car,” said founder Ranjeet Deshmukh, explaining how RoadBounce’s algorithm tracks vibrations.

Vibration-based models have a key advantage over the visual approach. “You need to have a line of sight to the problem to detect it visually, which may not always be possible when there is water or snow on the road,” Mr Deshmukh told The Straits Times.

RoadBounce’s geotagged reports are reinforced with AI-based analysis of visuals of the problems detected to help classify them accurately. If needed, the data is also put through human checks.

Since launching its first road-monitoring product in 2020, RoadBounce has covered more than 65,000km of roads in India. It has even worked for Singapore’s Land Transport Authority, developing a custom-built robotic vehicle to ensure roads being relaid meet necessary prescribed smoothness levels.

Road bumps to mass adoption

In October, the National Highways Authority of India announced it would deploy survey vehicles in 23 states covering 20,933km of roads to collect data on defects that would be analysed using AI.

Compared with manual inspection of roads, AI-powered tools are not just cheaper, but also far more efficient.

Mr Gupta said running a Nayan device for a year, including the submission of AI analysis reports to a client, costs around 40,000 rupees (S$570), which is far lower than engaging a human for the same task.

These tools also eliminate potential biases, such as a public works department employee who may not want to flag a pothole or a crack on a stretch of road that has been assigned to him for upkeep.

“That’s why we came up with a proactive mechanism, because AI is something that does not discriminate, and it’s always on duty,” Mr Gupta told ST.

But these AI-enabled tools are only as good as the human desire and capability to act on the data. No amount of technology can eradicate potholes from Indian roads if the city authorities remain lackadaisical about fixing problems flagged by these tools.

The experience with Indian cities has been mixed, with some prompt to act and some not so. Mr Gupta said Ahmedabad, the biggest city in the western Indian state of Gujarat, has been one of the better performers, with nine out of 10 flaws reported on Nayan’s devices repaired by the local authorities.

There are other Indian cities, which he did not name, where this repair rate has been five or even less.

Ahmedabad conducted two pilot drives with Nayan devices mounted on ride-hailing cabs and public buses in 2022 and 2023.

Mr Ramyakumar Bhatt, deputy municipal commissioner at the Ahmedabad Municipal Corporation, said that each time a problem was flagged, it was assigned to a specific person in charge of that area to ensure action.

“The beauty about this AI-based technology is that it compels action,” he added, referring to how an error would continue to be flagged each time a bus passed that particular spot until it had been repaired.

Encouraged by the results, the city in December issued a tender to deploy 250 AI-enabled devices to monitor its roads over the next five years.