Can ride hailing businesses scale sustainably when relying on LIDAR for autonomous driving?

The success of a ride hailing business lies in its ability to not only acquire and maintain customers but also scale and adapt to changing needs. Although autonomous driving promises to change the profitability of ride hailing business model by replacing the bottleneck which is the driver, this article will try to analyze if Lidar can act as the new bottleneck and also provide alternative approaches to scale fast.

 A lidar can cost between $7000 to $80,000. If one considers a sensor portfolio similar to that of Waymo which consists of 5 Lidars, 4 radars and 29 cameras, it will cost around $35000. Considering it is deployed in a city like London, which needs around 40,000 ride hailing cars, it would cost around $1.4 billion just to deploy fleets of autonomous vehicles in one city. This would make scaling a huge challenge for ride hailing service providers. Although this problem exists today, it would not be a bottleneck in the next 2 years. What remains a challenge is to pick the right option among the plethora of alternatives available.

1. Substantial drop in price of Lidar: The price of the lidar can drop substantially if there is enough production volume. Although the current price of Lidar is around few thousand dollars, due to the increased production, the Iris lidar range by Luminar will cost around 500$ to 1000$ while the Lidar by Velodyne will cost around 100$. This price makes it possible to scale quite conveniently but comes with a huge disadvantage of late market entry since it will be available from 2021 or 2022. A firm which waits for Lidar prices to drop would have already lost a significant market share. There is also an unforeseeable risk which could reverse drop in prices for e.g. consolidation of the Lidar suppliers.

Furthermore, Lidar has two operational challenges: Firstly, the range is dynamic, which means automated systems cannot always rely on the Lidar to give a complete picture of its surroundings. Secondly, most of the Lidar based automated systems today rely on HD maps which makes the process of scaling quite strenuous.

 2. Solid state Lidar: InnovizOne, a solid state Lidar will cost a few hundred dollars and promises to go in production by 2021. Since a solid state Lidar doesn’t have any moving parts and costs less than a Lidar, it is one of the promising alternatives. Infact BMW has a partnership with Innoviz where it will use its solid state Lidar for building its first generation of self-driving cars.

The problems that solid state Lidar will face is unpredictability. It is very difficult to predict if lidar is the best way to achieve a SAE level 5 autonomy. Andrej Karpathy, the Tesla Autopilot Lead, holds the view that Lidar is a shortcut and avoids the fundamental problems of visual recognition.

 3. V2X connectivity: Apollo robotaxi, a partnership of Baidu and FAW group has over the horizon perception, which enables it to view at a distance greater than 250 meters. This is a great example of how V2X connectivity can replace Lidar. It achieves this by a network of roadside units. These roadside units help in mapping the road traffic and broadcasts this information to the ongoing cars. Furthermore, Roadside units can also provide detailed real time 3d maps which can potentially reduce the number of sensors needed in an autonomous vehicle while providing a better picture of the road ahead.

One of the biggest limiting factors of V2X connectivity is the need for huge upfront investment in infrastructure. This huge support needed can act as a deterrent for adoption of V2X. However this data can be shared by more than one ride hailing partners and hence, the cost of setting up the infrastructure can be recovered.

4. Advanced radar systems: The automotive radar developed by UHNDER can detect cyclists, people and many other objects, at much better resolutions than ever before. This means that a system of radars, cameras and few low costs sensors is enough to achieve full autonomy. This not only reduces the cost substantially but also the amount of processing needed to reach a decision. Furthermore, Hydranet, an ensemble neural network which is being developed by Tesla could make level 4 autonomy a reality through a series of cameras and radar system. This model is particularly interesting since Tesla has already integrated their cars with cameras and other systems, allowing it to capture and test data on a daily basis. If successful, Tesla can deploy level 4 autonomy with a simple OTA update.

Sustainable solution

Of course, there are other promising solutions such as a set Stereo Cameras or normal cameras which are getting closer to Lidar based systems on the KITTI vision benchmark with the potential to replace LIDAR. The architecture that the firm chooses will decide its fate. While emerging systems which are cost effective and accurate are still under development and will need a year or two for deployment, the existing ones are more costly to scale but can get first mover advantage and reap network effect. Since the switching cost for a rider is quite low, it is crucial to build a system that has low variable cost and can scale quickly to get first mover advantage.

This is why V2X connectivity and advanced radar system could be one of the best ways to scale since both approaches have lower variable costs and can be easily deployed on a larger scale.