Demystifying Lidar Pricing — Part 1
The cost of lidar sensors has been a hot topic for years. In this short series, we’ll attempt to provide some useful context to understand lidar pricing and the factors that impact it the most. Part 1 includes a discussion of price and its relationship to performance.
Recently, Sabbir Rangwala wrote an interesting article at Forbes — “Money for Somethin?” — in which he made a great attempt to quantitatively evaluate pricing claims from various lidar suppliers.
Using the range (R) and points per second (PPS) specifications of the lidars, Rangwala drew some preliminary inferences on expected pricing for various lidars (“projected price”) and compared those to prices publicly communicated by the corresponding lidar suppliers (“announced price”). In a few cases, he observed that the announced price is much lower than the projected price and then discussed reasons why that might be the case — such as: software vs. hardware mix, innovations in system design, scaling of bill-of-material (BOM) costs, and quality/manufacturability.
Since lidar is a relatively new technology in many markets, the topic of pricing is worth a deeper discussion. In his article, Rangwala uses a price scale factor to inform his analysis. Since we don’t know how the factor is calculated, we can’t specifically comment on his projected numbers. However, we can make a few pertinent observations to expand upon his commentary.
Normalizing PPS
Rangwala’s approach to analyze pricing rests on two key performance indicators (KPIs) for lidars:
· R — maximum distance that the lidar can “see” an object of a certain reflectivity
· PPS — how many points (or pixels) the lidar can generate in an image every second
These are indeed two very important KPIs to consider in establishing a price for the lidar — hence, these provide a very good starting point. That said, a more complete analysis would have to take into account other KPIs as well.
Consider, for instance, two different lidars that have a similar R and PPS but different fields of view (FOVs). When we compare the performance of these two lidars within a common FOV, we would notice a difference in the point densities (pixel densities) of the two lidars — or equivalently, a difference in their angular or spatial resolutions within that common FOV. Angular resolution within a given FOV is an important KPI because it helps us more easily determine the difference in capabilities of two different lidars when it comes to object detection (“seeing”) and classification (“recognition”).
To understand this better, lets’ start with the image below, taken with a Cepton Helius™ Smart Lidar system. This system includes our Helius perception software running on an edge computer connected to Vista®-P series lidars. Each box in the image corresponds to a detected object that is moving. The color of each box tells us what type of object it is — e.g., green for pedestrian and purple for vehicle. The more points within the box, the more accurate the lidar perception system is likely to be in detecting and classifying each object correctly.
We can quantify the importance of this by using a specific example from Rangwala’s article. One of the lidars cited by him is capable of ~650,000 PPS and ~120 m range on a 10% reflective object (10% being a typical industry reference for assessing range performance on a dark-colored, weakly reflective object). This reference lidar — a Cepton competitor in certain markets — also has a 360° horizontal FOV (HFOV). In comparison, one of Cepton’s most popular lidars — the Vista-P60 — has a similar range but with an HFOV of 60°, and it generates ~315,000 PPS (single return mode). The reference lidar is well suited for an application which requires a full 360° HFOV centered around the lidar. However, what happens when the application does not require a 360° HFOV? Indeed, numerous applications where lidar is used do not require 360° HFOV and many of them can very effectively work with HFOVs ranging from 60°- 90°.
In the 60° HFOV case, we can compare the PPS between the reference 360° HFOV lidar and the Cepton Vista-P60 or Vista-P90. As shown in the chart below, the reference lidar would have a PPS (within 60°) ~ (60°/360°) x 650,000 or approximately ~108,000 PPS, in comparison to the Vista-P60’s ~315,000 PPS. Similarly, the PPS (within 90°) for the reference lidar would be ~ 162,500 PPS, compared to a PPS (within 90°) of ~315,000 PPS for the Vista-P90. In other words, the Vista-P60 offers ~3X PPS within a comparable HFOV and the Vista-P90 offers ~2X PPS within a comparable HFoV vs. the reference lidar, thereby enabling much higher accuracy in object detection and classification.
From Point Clouds to Perception
At the same image refresh rate, the higher PPS multiple translates into a substantially higher number of points on a given object at the same range (R). This leads to much higher accuracy in detecting and classifying objects. Moreover, there is a difference between a simple notion of range (R) of a lidar and the range at which the lidar’s perception software can detect something in the environment as an object (Rdet). To a lidar perception algorithm an object is a cluster of points — typically a few points within a certain distance of each other. So, when we try to detect objects (rather than single points) in the surroundings, an R of 120 m could become much lower in practice (i.e., Rdet < R) because object detection algorithms need a certain minimum number of points to identify something (or someone) as a real object. In a situation like this, two sensors with identically specified ranges (R) might actually provide very different detection (Rdet) and classification (Rcl) range performance for certain objects, depending on the PPS within the specified FOV.
What we’ve found is that many of our customers — in markets ranging from AV/ADAS to Intelligent Transportation Systems (ITS) to Smart Spaces (e.g., Crowd Analytics and Security) — hugely value this PPS multiple. So, a sensor (Vista-P60) that superficially seems to have a lower PPS and lower HFOV compared to a 360° HFOV sensor, in fact turns out to be far better in solving difficult detection and classification problems in several industries. This is one reason why the Vista-P60 is best in class for range and resolution in the affordable price band for lidars — and is unmatched in the industry for what it can offer for a variety of applications, including those which require full area coverage (e.g., buildings, airports, traffic intersections, malls, and so on).
Taking all of that into account, when we consider pricing, 360° HFOV lidars with directional PPS (within a specified FOV) that come close to that of a Vista-P60 or a Vista-P90 are typically priced multiples higher. This makes the Cepton sensors ideally suited for many applications and also provide us a much easier path to price reduction when we look at scaling the technology for higher performance at very high volume (more on this in a future post). There are other performance factors that do come into play when considering pricing for specific applications, but those are typically secondary to the ones discussed above.
Read the following articles in this series: Part 2 | Part 3
____________________
By Dr. T. R. Ramachandran, Chief Marketing Officer at Cepton
Follow Cepton for more information: LinkedIn | Twitter | YouTube
