Can India’s ADAS mandates succeed in India’s CV segment?

The same alert cannot serve every driver

This is where I think the industry needs to move past treating ADAS as one undifferentiated stack bolted onto every vehicle category. A city bus, a long-haul truck and a HazMat tanker are not variations of the same problem. They are three different problems.

Take a bus driver navigating dense city traffic. Their risk universe is dominated by vulnerable road users: pedestrians stepping off a crowded platform, cyclists sharing the same lane, a two-wheeler emerging from a blind spot exactly as the bus begins a turn. The intelligence this driver needs is spatial and immediate: which side is the threat approaching from, how much time is left to react, and how urgently does this need to be surfaced without adding to an already high cognitive load in stop-start traffic. Here, the value of the system lies in resolving ambiguity in the moment, distinguishing a pedestrian who is crossing from one who is merely standing near the kerb, and doing it fast enough, and clearly enough, that the driver can act.

Now take a long-haul truck driver on a national highway at 2 a.m. The frontal road may be completely clear. There may be nothing to detect in the classic sense no pedestrian, no vehicle cutting in, no lane drift yet. And still, this is often the more dangerous moment, because the risk isn’t external, it’s internal. It’s fatigue building over hours of monotonous driving, a driver whose eyelids are drooping even as the hands stay on the wheel. A system tuned only to detect external hazards will stay silent right up to the point of failure. This is precisely why Driver Drowsiness Detection has to function as its own vigilant layer, correlated with time-on-road, blink rate and head-pose not as an afterthought bolted onto a forward-collision system.

And even within “truck” as a category, the physics change constantly. A tanker carrying liquid cargo has a completely different centre of gravity and rollover threshold than a flatbed carrying a fixed load. A 12-metre rigid truck brakes and turns differently from a multi-axle articulated one. Load type changes stopping distance. Vehicle length changes blind-spot geometry and the swept path in a turn. A braking decision that is correct for an empty container truck can be the wrong decision too aggressive, too late, or simply mis calibrated for a fully loaded tanker taking the same corner. Any system that issues an emergency braking recommendation without accounting for these variables is applying a generic answer to a vehicle-specific question.

Perception without context is just detection

This is the distinction I keep coming back to: detecting an object is not the same as understanding a situation. A commercially viable ADAS stack has to fuse what the camera sees with what the vehicle is its class, its load, its length, its dynamics and with what the driver’s own state is at that moment. Only then can it decide whether a given event deserves a gentle visual cue, an urgent audible alert, or active intervention.

This is also why ADAS systems need to be developed and trained on data that reflects the realities of Indian roads, rather than relying primarily on models built around more structured and predictable traffic environments. Indian driving conditions vary significantly across cities, highways and rural roads, as well as across different times of day, weather conditions, road users and vehicle types. Factors such as inconsistent lane markings, mixed traffic, two-wheelers, pedestrians, varying road infrastructure and unpredictable manoeuvres create scenarios that conventional datasets may not fully capture. Building systems that can operate reliably in this environment requires extensive real-world Indian driving data and the ability to interpret not just individual objects, but the context in which they appear. The objective is not simply to detect what is on the road, but to understand what it means in that particular driving situation.

From passing a test to earning trust

Regulatory validation, whether AIS184 with ARAI in India or EU2144/EURO NCAP protocols in Europe, remains essential. It is the shared language the industry needs to prove baseline reliability, and we’ve built and validated our own systems against exactly these standards. But validation against a standard test suite is a necessary condition for deployment, not a sufficient one for safety.

The deeper test happens on the actual road, across actual fleets, over actual months of driving. It’s why the metric that matters most to us isn’t a pass/fail certificate; it’s whether high-risk driving behaviour measurably drops once a system is deployed at scale, and whether drivers keep trusting the alerts long after the novelty wears off.

India’s roads are not a harder version of the environments ADAS was first designed for. They are a distinct operating domain, with their own vehicle mix, their own traffic behaviour and their own risk patterns. Meeting a mandate is the entry ticket. Earning the trust of the driver behind the wheels of a bus, truck or tanker, city road or highway, day or night, is the actual work. That’s the bar we should be building to.

(The views expressed by the author, Nisarg Pandya, in the above article are personal)