Elon Musk’s FSD vs Rest of the World’s LIDAR
Even old-school automobile makers such as Mercedes-Benz, Audi, and BMW have started toying with their own ideas of driverless vehicles.
In 1990, I watched the 80’s movie “Back to the Future, Part 2”. In an era when cars used to travel only on roads, the film depicted flying cars around an airborne New York taxi stand in 2000 (as ridiculous as it may sound now).
Almost 35 years have passed since the release of that movie, and while cars are yet to soar above us, the next significant milestone for humanity will no doubt be driving without human drivers.
What? Is producing driverless cars that difficult?
The biggest impediment to driverless cars is the environment. Whether it’s the chaos on the roads in the Indian subcontinent or the discipline of Singapore’s traffic, it’s the human element that is unpredictable. For a driverless car to be successful, we expect it to drive perfectly, making absolutely no mistakes on the road that could result in fatalities.
The fact is that when a human drives, all senses at his/her disposal are utilized, along with the brilliant mind that is able to process signals in a fraction of a second. As we architected systems to perform the same task, we became aware of how challenging it was for an algorithm in this case to accurately mimic human behavior.
That said, companies are racing ahead to produce driverless cars. Although the technology is available, as we have seen previously, it is the edge cases that are actually slowing down large-scale adoption.
What do I mean by edge case?
There will always be a specific scenario on the road that the AI-powered software hasn’t trained on due to a lack of exposure to the data. The more data the system collects on driving, the more accurate it becomes in making road predictions.
Presently there are 2 technologies battling it out for superiority in this space.
Geofencing
The term itself accurately describes the act. We use training data specific to a particular geography when building such self-automated systems. This acknowledges that there are certain social and even cultural undertones that impact driving habits in a location. As a result, we geo-fence these vehicles so that they only operate in certain regions, not all.
Sensors play a key role in geofencing.
A lot of sensors are used here, which include LIDAR (Light Detection and Ranging), radar, and even cameras. The extended use of sensors makes data collection a lot more extensive. Obviously, the cost of such an approach would increase, pushing manufacturing costs higher.
So what is LIDAR?
LIDAR is a technology that emits laser light into its surroundings, then gathers measurements from the sensors of all objects around it and creates a dense 3D point cloud representation of the scanned area.
If we have LIDAR, why do we need radar?
LIDAR isn’t really effective at long distances or when the climate in the surrounding area is poor. Radar steps in to fill in the gaps in the data.
Other fields, including atmospheric sciences, have previously used remote sensing technology like LIDAR for mapping land terrain. Self-driving cars seemed like a natural application of a proven technology.
Companies using this technology, including Waymo, Mercedes, Audi, and many more, are testing their prototypes. In fact, Waymo has been able to move the fastest and now has its fleet of robotaxis deployed in the Phoenix, Arizona, metropolitan area.
The fact that Waymo has already been able to achieve Level 4 in autonomous driving is an achievement in itself.
More about these levels later.
Tesla’s FSD
You can trust Elon Musk to break away from the pack when he wants to. We can see the way he juggled Twitter in the air until it assumed the form he desired.
To start with he didn’t adopt geofencing. The technology they built (FSD — Full Self Driving) needed to access a wider base of users across the globe so they decided to take a very different route.
So what is FSD and how is it different from LIDAR?
FSD technology uses only cameras to capture images of the objects. It doesn’t use other remote sensing equipment such as LIDAR. By limiting this to cameras only the cost overall is reduced.
They decided against adopting LIDAR, in fact, Elon Musk has come out in the open against LIDAR calling it “a fool's errand”.
How are they making up for this?
Their secret sauce is training data. Tesla will deploy their FSD-powered vehicles in the field at the earliest in order to gather as much data about the field as possible. The more data they gather about the field, the more they can train their autonomous Neural network-based system to be more ready for all possible scenarios they may encounter in prime-time
You can almost certainly guess which is why Tesla’s vehicles have a larger number of accidents.
As of now Tesla’s FSD has achieved Level 3 in Autonomous driving but Elon says they could get to Level 4 sometime this year.
Now, what are all these different levels of autonomous driving?
This article below provides a clear explanation of the various levels of autonomy, where 0 represents no automation (human driving) and 5 represents complete autonomy.
This scenario reminds me of the CDMA vs. GSM telco war decades ago, when telecom as an industry was evolving and giants had to pick sides.
Then suddenly, a new player entered the fray and raised the stakes.
BYD, Tesla’s largest Chinese rival, has strongly supported LIDAR. On top of this, their EV vehicles are selling like hotcakes throughout Asia. Their cars come at a considerable discount to Tesla’s, making them formidable competitors in certain markets (thanks to Trump’s taxes).
Even old-school automobile makers such as Mercedes-Benz, Audi, and BMW have started toying with their own ideas of driverless vehicles.
It looks that we could be driving autonomously much sooner than cars can fly.
