[OffRoadFun]

Mod Edit: We have received clarification from reliable sources that this is only a collision warning system. No data is being collected in the form of videos or images.

I just saw a Mobileye (An Intel division) testing in Bangalore, on Outer Ring Road. Not many may know but Tesla initially worked in collaboration with Mobileye's camera based technology and later fell apart as Mobileye felt Elon was too reckless with rolling out FSD, too early, compromising safety.

They recently demoed self drive in notorious NY city traffic, only company to do so. In short, in my opinion, they are one of the most advanced.
The car did not have lidar or cameras visible though they had mounts to install them in future. Seems like initial data collection about roads etc for India. May not be full AV, but they can sell partial autonomous driving features in India may be in future. Thoughts?

[abhishek46]

Autonomous vehicle rely quite a bit on road infrastructures, like lane markings, signage etc.
Till the time road infrastructure is completely developed, Autonomous technology will have a hard time here.

Secondly, there are 'unique' scenarios here, like random animals on the road, extra large moon-crater potholes (sometimes filled with water), random people jumping in front of the vehicle, extremely close bumper to bumper tail-gating.
Can Autonomous systems account for so many of the above random events?

[sreeharipv]

Talk about accelerated learning. I always thought if machine learning is really a thing and neural nets are mimicking human neuro system, why Tesla shouldn't be doing all their 'learning' in India so that it'll be a piece of cake back in the US.

[abtyagi]

While my time in Zoomcar, we did add some MobileEye devices on some of the cars, IIRC 20 odd cars. It wasn't much effective in Indian road conditions. Hopefully being trained here in real traffic shall improve the efficacy.

[RedTerrano]

Any such software will work using three basic steps.
Sense.
Analyze.
React.

Besides this, as it has already been pointed out, it will also rely on infra (well demarcated lanes, road signage etc) as well as human discipline (keeping proper distance between vehicles, using zebra crossings etc).

Sadly, most of the things are lacking in our country.
Considering all this, I feel we have a long long way to go for any kind of warning system to reliably work.

The mention of animals on our road reminded me of a decades old sad incident in Kolhapur.
One of my acquaintances was riding pillion behind his best friend on his bike. They were on the road leading to the river. Plying the same road, was a herd of buffaloes who were being taken for a bath.

These two were merrily yapping away, riding down the road, ignorant of the too familiar sight of buffaloes. Suddenly, one buffalo turned her head and the horn pierced the rider in the chest. Fortunately it was instantaneous and he did not suffer. He died on the spot.
My acquaintance on the other hand was so shaken up that he developed high fever and needed to be hospitalised. He has never ever travelled down that road since.

[Steelneo]

Originally Posted by abhishek46

Autonomous vehicle rely quite a bit on road infrastructures, like lane markings, signage etc.
Till the time road infrastructure is completely developed, Autonomous technology will have a hard time here.

Secondly, there are 'unique' scenarios here, like random animals on the road, extra large moon-crater potholes (sometimes filled with water), random people jumping in front of the vehicle, extremely close bumper to bumper tail-gating.
Can Autonomous systems account for so many of the above random events?

True.

For every possible scenario's that are realistic i think a model can be made and tested. The things that are in Government control are the one which will be difficult to predict . It required repeated testing iterations.
Another important thing is people's disciple to road traffic rules & the two-wheeler hero's . That will be a challenge here.

Above all if any company that can achieve 40% accuracy at present date, then achieve above 80% in western countries. Hope we will change by the time and follow rules personally.

Mod Note : Post Edited. Smiley usage is restricted to two per post on Team-BHP. Please do NOT use more than 2 smileys in a post.

We advise you to read the Forum Rules before proceeding any further.

[Gaur]

Originally Posted by abhishek46

Secondly, there are 'unique' scenarios here, like random animals on the road, extra large moon-crater potholes (sometimes filled with water), random people jumping in front of the vehicle, extremely close bumper to bumper tail-gating.
Can Autonomous systems account for so many of the above random events?

True,there are multiple scenarios that still need to be accounted for.
My better half's work profile is somewhat related to this and the various scenarios they encounter are funny and scary at the same time.

what would you make of it when your vehicle confuses moon for a traffic light as below.

https://indianexpress.com/article/tr...light-7424808/

[OffRoadFun]

Mobileye has released a video about learnings in their computer vision algos from driving in India. talks about various camouflaged "obstacles" and other unique scenarios.

The video is impressive in the sense their computer vision algorithms are mostly able to identify "objects" and their boundaries with high level of accuracy. How they will use it still needs to be seen, but the demo is indeed impressive. I have always believed ADAS may not for Indian traffic, but this video forces me to re-think that assumption, at least for partial features and collision warning:

https://www.youtube.com/watch?v=xL7i2MQHORY

[Vivek Menon]

I think it would be interesting to see how Mobileye would work in Indian conditions. Now that they have partnered with Mahindra for XUV 700, we can get the know the effectiveness in real life scenario soon. Having used vehicles installed with Mobileye software and having worked in a team installing Mobileye after market system in buses, I can safely say that system works super good even after many years of installing.