Self-Driving Technology: Lidar VS Camera
Hi, I’m Viram from Vested and today we’re going to talk about Self Driving Technology. We’ll first discuss what exactly is this technology and who are the key players in this space. Then, we’ll talk about the two opposing approaches the players are taking to build this self-driving technology, we’ll compare its pros and cons, and then lastly we will find out who is it that will emerge as a likely winner in this self-driving race.
So, let’s dive right into it.
What is self-driving technology
So, what exactly is self-driving technology?
Well the definition says that a self-driving or driverless car is a vehicle that uses a combination of sensors, cameras, radar and artificial intelligence to travel between destinations without any human intervention. Cool right?
Now, are all self-driving vehicles the same? Not really.
There are actually 5 levels of automation as defined by the Society of Automotive Engineering, and the higher the level of self-driving designation, the lesser the human intervention is required.
The first 3 levels have a high to medium level of human involvement, while the last two have little to no human involvement in the entire process of driving a car. Players like Cruise, Waymo, Uber, Lyft, Tesla, Aurora and Intel are all either working on building this technology or have done so in the past.
Currently, however none of the players have achieved the highest level which needs absolutely no human intervention in the entire driving process. It is also unclear when the high level can be achieved and what time period will we get there. Companies in this field have been underestimating the time it takes to actually achieve this large key milestone.
In 2018, Waymo prepared to order 82,000 self-driving vehicles. But most recently in 2021, it has been known to have about only 600 cars in its pilot fleet. A big difference.
Uber thought that by 2020, it would have 100,000 self-driving vehicles. But currently it in fact has none. Because it sold its self-driving division to this other player called Aurora Technologies.
Similarly, Lyft thought that by 2021, the “majority” of its rides would be done autonomously completely without a driver.
Turns out, getting to the highest levels of automation is much much harder than anyone ever thought so.
There are two approaches to achieving this highest level of automation:
The first is the evolutionary approach and the second is the straight to full autonomy approach.
Now, we are going to be looking into what exactly both of these approaches mean.
The evolutionary approach
The evolutionary approach is a strategy where you slowly level-up one after the other towards the highest level of automation.
This approach uses multiple off-the-shelf technologies such as radar, camera and sonar.
By using such off-the-shelf technologies or ready technologies, companies using this strategy can deploy advanced driver-assistance systems, also known as ADAS into existing vehicles and evolve their capabilities to achieve greater levels of automation over time. This approach is great because companies can continue to de-risk the technology, while at the same time getting cash flow from selling these ADAS systems to existing cars. This is the strategy that essentially Tesla, Mobileye and several others are pursuing.
Full autonomy approach
In contrast, the second approach is to develop the necessary technology called lidar to jump straight to the highest level of automation. This is the approach that Waymo, Cruise and Aurora are taking. These companies are able to afford large amounts of research and development budgets because they are either getting large amounts of funding from private investors or they are going public via SPAC.
These companies generally do not have products in the market yet, so they rely on investment dollars to essentially continue funding their development and that development which is actually taking more longer than expected.
The key difference between both of these camps is the answer to the question: Do you need advanced technology to achieve this highest form of automation? The companies employing approach 2 tend to believe so. They don’t believe that evolving ADAS which is the driver-assisted systems will ever be able to sufficiently achieve the highest level of automation.
Whereas, the other camp which is taking the evolutionary approach believes that you don’t need lidar, the first approach is much more efficient in the long run. And that’s how they create the highest level of automation.
Comparing technology behind these approaches
Now, let’s spend some time comparing the actual technologies behind these two approaches.
So, what’s this technology that Waymo, Cruise and Aurora want to build? It’s essentially called lidar. Lidar is basically light based radar. It shoots lasers around the vehicle and as light is reflected back to the device by the surroundings, the vehicle creates a high fidelity map of those surroundings. It can actually see.
In 2015, the cost for a lidar was US$75,000 per device. Since then, that has gone down to $500 per device. And despite the continued decline in costs, lidar is still too expensive for mass deployment which means it is not scalable.
Despite Lidar’s costs, proponents of this approach believe that this is the only way to achieve high scale automation that needs minimum human intervention.
Characteristics of lidar approach
Here are key characteristics of this approach:
Lidar captures the environment at a much higher detail than a vision based camera system, which is our second technology.
The higher detail captured from the environment is then fused with other sensors to create a HD map and to train a machine learning algorithm that drives the car.
So, what’s the downside? Well, this machine learning model requires a lot of training data. This means that you have to deploy a very large vehicle fleet equipped with lidars to drive through all the different cities and under different driving conditions to get that data. And this is too expensive to achieve realistically.
While the other option is to use a lot of simulated driving to train machine learning models. In fact, Aurora was doing this earlier. It was able to simulate 2.25 million left hand turns before actually testing that capability on public roads.
Now, let’s look at the second technology, which is vision only approach.
The folks who support this system, are trying to use vision based systems. The most advanced on this front is none other than Tesla. It builds the car, the sensing technology, the data, and the machine learning models all by itself. Other carmakers partner with external safety & assistance providers like Mobileye, which is owned by Intel.
Recently, Tesla’s system got rid of the radar and now solely relies on cameras. Cameras used in this vision system cost less than US$100 for the entire car, and each Tesla is equipped with eight cameras.
Characteristics of vision only approach
So, here are the characteristics of the vision only system.
First is that, there are like cameras you find in your smartphones, which means they are cheap. And because they are much lower costs, you can deploy them at a massive scale.
Massive deployment means that even though the surrounding data generated is of lower resolution than that of lidar, a vision based system a vision based system can overcome the resolution limitation by training the artificial intelligence or the neural network with a much larger data set.
Also, because the system cost is much lower, it can be deployed in shadow mode in production cars today. In shadow mode, the system records the driving data, lets the human drive, and then compares what the artificial intelligence model would have done in the same situation. This transforms the problem into what is called a supervised learning problem, where the user not only pays for the privilege of acquiring the car, but also helps train the system better! Tesla today has millions of cars globally training its machine learning model.
Although it’s still unclear which strategy will prevail, what history tells us is that achieving the highest form of automation may take longer than expected. As an industry, the revenue in this industry is pretty much non-existent! The projected industry revenue is less than US$9 billion by 2030. That’s even less than 1% of Amazon’s current quarterly revenue!
The two technology approaches are converging towards the same level 5 high-automation future. And the folks at the lidar camp are racing to push the cost down as fast as possible, so that they can deploy more and more vehicles to get more real-time world coverage. Whereas the folks in the vision based camp are racing to the ladder up to the highest level of automation by making their machine learning models more and more smarter everyday.
In the past, any time there were two competing standards or approaches for the same application, the winner wiped out the losers. The winners of these standard wars gained massive adoption rapidly, while the losers disappeared.
For the lidar system to prevail, two things must happen. The first is that it has to achieve the highest form of automation and the second, is that it has to be at the same cost or cheaper than the camera based system.
Meanwhile, for the vision based system to prevail, only one thing has to happen which is, it has to just achieve the highest level of automation. Because the other issue, the camera prices has actually been solved because of the smartphones. The cameras are priced very low currently.
In conclusion, it’s still unclear if vision based alone can achieve that golden level of automation. It is undeniable that the vision based strategy has the cash flow and data advantage though vs the other.
So, lastly in today’s we discussed what self-driving technology is, what are the key players the likes of Tesla, Waymo, Cruise, Lyft – what are they doing. We also discussed what are the technologies that players are using to create fully autonomous vehicle and what needs to be done to get to a future where we can actually cruise around in cars without anybody driving them.
Hope this was helpful. Stay tuned for next week!