LiDAR: what the future looks like for this sensor
Internet archive is full of great articles on this topic: how Lidar works, what are its key advantages and disadvantages, what are the most popular models, etc. In this brief article, I’d like to reflect on what’s been happening in this space to understand what role Lidar will play in the AV future.
Those of you who are familiar with the topic, know very well about polar opinions on the usefulness of Lidars in self-driving cars. While representing Tesla’s vision towards an autonomous future, Elon Mask and Andrej Karpathy fully rely on cameras and deep learning algorithms. Elon Musk is a strong opponent of Lidars. Last year, at Tesla Autonomy Day he said:
“Anyone relying on lidar is doomed. Doomed!”
The two key reasons of his view are that Lidars are expensive (still true) and unnecessary (not really). In his opinion, while Tesla’s approach is not the easiest one, it’s the right way to the adoption of self-driving technology at scale. ScaleAI has a good piece of analysis on this topic saying that while Tesla’s approach is possible, it’s still a way to go.
At the same time, the majority of AV companies actively use Lidar among other sensors. Given the current state of technology and relentless focus on safety, it’s a no brainer decision. Oliver Cameron, CEO at Voyage, stresses that they are not going to focus on optimizing costs today but rather optimize for performance to get to truly autonomous driving sooner than others. Even though he said it before Covid-19, Lidar plays an important role in AV’s perception systems to speed up commercial deployment.
This sounds like a plausible logic in light of an assumption that we will eventually get to a point when Lidar’s price is no longer a show stopper.
Let’s shed some light on Lidar’s evolution from a few angles. Time and money invested in any technology are among the two biggest drivers affecting its development speed. The graph above shows a clear upward-looking trendline of VC money invested in Lidar startups. This gives confidence that increased competition in the space will eventually lead to lower prices and improved product characteristics including range, size, and robustness.
One thing is sure, Lidar characteristics are indeed getting better across the board: some companies work on solid-state Lidars, the range increases and the sensor itself gets smaller and smaller.
The price is on top of everyone else mind as well. Hence it’s a matter of time when we see a drop. Recently, we’ve seen internal LiDAR projects to emerge or deals when AV companies acquire LiDAR startups to control costs and get exclusivity on best-performing sensors. Waymo has even started to sell its Lidar to customers last year.
However, one of the boldest statements belongs to Luminar. It is aiming to reduce the cost to around $1,000 a unit for autonomous applications compared to Velodyne’s Lidar, which can cost as much as $75,000.
Let’s sum up what we know and infer about Lidar:
- It’s expensive these days, but its price is likely to go down;
- Lidar is actively used in AVs as it complements the perception system;
- With time, as algorithms improve, dependence on Lidar might go down.
To have a clearer picture of what will happen to Lidar, let’s go back to the biggest self-driving technology promises. Among them, safety is the central one. Hence, in a not so distant future, in which Lidar price drops and self-driving technology penetration increases, the highly likely scenario is that no one is going to question using Lidar. This is especially true given efforts to fully realize Lidar's potential to extract semantic information.
So given the safety promise, it’s always better to have multiple sensors for situations when one sensor type is ‘blind’, so we can rely on the other one to get the missing information. In the coming future, AV companies will simply treat Lidar as they do a radar sensor today (that is cheap, reliable, useful), and use Lidar whenever it makes sense to ensure high safety.
