This is a continuation of my thoughts on where we are headed with driverless cars. I would recommend reading this series from part 1.
Previously we discussed how will driverless cars change our environment. In this post, I wanted to provide some my thoughts on who will have the biggest impact in this market.
There are a lot of players in this market – Google, Tesla, Uber, and the automotive manufacturers. IMHO, those with the biggest impact will need the following infrastructural items to win:
- The safest driving models
- High-density batteries
- Parking and car servicing network
- Efficient procurement of vehicles
- Efficient deployment vehicles to satisfy demand
- Partnerships with municipalities
- Labor and rural focused solutions
Autonomous driving safety is paramount. If rider safety is compromised, it will lead to mistrust in the system and more than likely to the transportation provider (as seen post the TWA crash). To ensure high-quality safety, autonomous driving models must have (1) enough data to train the models on, and (2) machine learning engineers and platform to train the models on
Getting enough data
As Elon Musk mentioned – one of the key differentiators is building trust in the driverless car technology is ensuring driver safety and the way to get there is by getting to ~100B miles (more miles = better-trained models the expected performance would be that driverless cars would be 10x more safe than humans). The more miles, the more data. The more data the better the training for the machine learning algorithms. The more training data, the better the execution in autonomous driving.
In that respect, Tesla is at a major advantage. With 200K cars on the road, Tesla is collecting a great deal of data on driver behavior and road conditions. They are logging all the road conditions at an alarming rate. Since October 2014, when they announced Autopilot, Tesla has logged over 1.3B miles of training data; which has led to a 40% (1.4X) improvement in safety. As the Model 3 comes to market Tesla’s ability to produce more vehicles the vehicle miles logged will only exponentially increase.
While Google doesn’t have as large of a fleet to capture training data, it is simulating data with Google Street View, and its ability to collect routing information from Google Maps and Waze. In order to catch up, Google is going to have to develop a strategy for gathering training data faster than Tesla.
Uber is slowly entering this race of data collection with its efforts in Pittsburgh and Phoenix, but it will need a lot more information.
Unfortunately, for the OEMs, they are too far behind in these respects to catch up. Their focus needs to be on the driving experiences riders are looking for.
One potential opportunity is for the OEMs is to forgo this capability and give it to Google and let Google be the OS, and the OEMs act as the vehicle manufacturer – adopting a similar model of Android’s relationship with all the cell phone OEMs. This means Google will often own the experience something the OEMs may not want to let go of. Additionally, because each make and model are different, Google may introduce far too many variables in its trained data set. Google needs to pick a standard vehicle for its fleet to train on.
On this note, regulators will need to manage the commissioning of any new make and model that enter the market, testing in more ways than just a crash test. We’ll address this in the next post of this blog series.
Machine Learning Engineers
While Tesla has a great method of collecting data, it doesn’t have the horsepower of Google when it comes to developing models. Google has been working on machine learning models since the first ads it served. The infrastructure and the teams its acquired and built have made it a force to be reckoned with. Their machine learning engineering has left the most valuable company in the dust when it comes to capabilities of their smartphone.
Tesla and Uber can leverage the great engineering talent of Silicon Valley, just as Apple has done, but it can’t keep up with Google’s machine learning muscle. That said, luckily for Tesla and others, much of the performance of machine learning models is not on the machine learning algorithms itself but on the data and the features built off that data.
While battery technology is important, I expect most vehicles to only operate <20% of the time (much better than the current <4%). As a result, most would suspect that the high kilowatt size of batteries may not be as important. However, with smaller vehicles (2 person vehicles with little to no trunk space), as well as long-distance hauls the need for high-density vehicles will be greater than ever. It is important that we’re able to pack as much kW into as little space as possible.
OEMs that can build highly reliable (cells don’t die often), high-density batteries can keep their cars on the road longer, ensuring more higher utilization – competitive advantage in this new market.
Parking and car servicing network
Additionally, these batteries will be the lifeblood of the entire ecosystem. if batteries are smaller but denser, they will require frequent charging, battery cells will need frequent replacements, vehicles will need to be cleaned and serviced during off-peak times, much of this will need to be coordinated via a parking and servicing network (discussed in part 3). Fleet operators that connect to this network will be at an advantage. Tesla is building towards that with its OEM supercharger network, but that is at a much smaller scale than what is needed for the complete switch. No other player has this network either. As mentioned in part 3 – this is a huge opportunity for someone to buy municipal parking assets and begin to cater to the near driverless future.
Efficient procurement of vehicles
Ride providers and asset owners will need to make sure they’ve purchased the right portfolio of assets. Based on the mix and design of the assets it will cater to different customer bases. In general, they will have to weight between two dimensions: vehicle design, and vehicle type:
Vehicle Design for consumer use will primarily focus on the economy and middle-tier use as it does today. Luxury options will continue to be an option. Sports cars and SUVs will be for recreational use, and not provided via the traditional fleet. The design will dictate function (e.g. a commute car may include desks and a coffee machine with a small trunk). Depending on the makeup of class of vehicle design it will lead a ride provider to cater to a specific type of rider. Most companies will go after the most mixed use vehicles as possible to enhance utilization.
Vehicle Type is the size and occupancy of the vehicle. I expect cars to get smaller (no need for hood space (no engine with electric motor vehicles), trunk space (addressed in the next paragraph), or seating capacity. In addition to vehicles not being fully utilized, car seats are typically not fully utilized. Since most riders will not own the vehicle itself they are not incentivized to purchase a vehicle that meets their maximum capacity needs, instead, they need a vehicle for their occupancy needs at the moment. Ride providers and asset owners will need to move OEMs to develop much more low occupancy vehicles (<2 people). There will be other uses for larger vehicles, I suspect the following mix of occupied vehicles:
- low occupancy: <2 people
- medium occupancy: 2-6 people
- high occupancy: 6-12 people
- mass occupancy: 12-50 people
Most if not all of those vehicles will only have space for some light baggage (~2-3 grocery bags per person). So what about the times where riders need trunk space or are hauling items in their pickup? Interestingly, because vehicles are autonomous they do not necessarily require someone with the hauled items (ie. you luggage on the way to the airport is in an unoccupied vehicle that is just behind you). If you need to hold onto the hauled items for some time (moving van), you can rent the idle time. Here is a mix of what these unoccupied hauled vehicles will be:
- No occupancy small haul (large car trunk)
- No occupancy medium haul (pickup truck)
- No occupancy large haul (medium size Uhaul/half a container)
- No occupancy commercial haul (container sized vehicle)
Special Note: Change to the environment
While we did not address this in part 4 of this series, primarily because we didn’t discuss the composition of the fleet at that point, it is important to note that with unoccupied hauling vehicles, there is a new opportunity for your vehicle to do the errands for you. Need to pick up dry cleaning, send a vehicle and have the cleaner load it in your car for you. Need to go pickup groceries or takeout, place your order ahead of time, and have the vehicle go grab it for you. Fleet operators become a transportation API to the world of in-person commerce. This is a huge opportunity to just focus on hauling of goods and services. Companies like Instacart, Postmates, Deliveroo and DoorDash are performing well here, but Uber with its entry into the foray as UberEats isn’t far behind.
Fleet operators will need to manage the mix of assets in their fleet to appropriately satisfy their customer base. One way to incentivize riders to use low demand vehicles is to change the economics to encourage more equitable utilization of the vehicles.
There is no clear leader here. No OEM is building like this, nor is any technology or ride-hailing company. Uber and Lyft might be the closest in understanding the rider needs on the type of vehicle, seeing they provide users with varying options and if a rider is pooling they know the number of passengers that are typical to help guide what the requirements will be for this future market.
Efficient deployment vehicles to satisfy demand
Not only do fleet operators need to have the right mix of vehicles, and the right location of an effective network of parking and services, they will also need to deploy their assets throughout a municipal effectively and efficiently to meet rider demand. Long wait times will easily cause riders to switch to another service. Ride-hailing companies such as Uber and Lyft have an advantage here as they’ve already begun to understand the driving behavior or ride-sharing routes.
Partnerships with municipalities
In addition to deploying vehicles appropriately, fleet operators will need to work well with municipalities to (1) ensure there is a smooth transition to driverless technologies, (2) build the infrastructure to cater towards driverless technologies (3) provide city services such as mass/subsidized transit.
In all three of these categories, OEMs have a much stronger relationship when it comes to transportation and government relations than the technology companies, ride-hailing companies, or even Tesla do. That said, it does make sense to dive a little deeper on each.
Google has spearheaded this the most as they’ve been working with cities and states throughout the country to get approval for their driverless car technologies. Being the one who helps get the exemptions passed opens the flood gates for not just Google but others, but as architects of the initial rules, it sets a foundation for what serves Google the best, and perhaps not the other players in this new market.
Infrastructure design building
As driverless technologies continue to become more mainstream, it will begin to have an effect on the design of our cities (see part 3 for my thoughts on this). Players in this market will need to help design and determine the infrastructure of how cities should handle potholes, streetlights, construction areas, and lane lines. Those who are actively working with municipalities will have an edge in this space. While no company is leading here, one company is falling quite behind – Uber. Their “ask for forgiveness, not permission” relationships with governments can lead to a lot of animosity between them and governments. They would be better suited to build a stronger relationship with the cities they operate in.
As driverless technologies begin to take hold, it is important for fleet operators to build vehicles that meet the needs of the city. Uber, Lyft, and Google are not looking at this at all (with the exception of Altamonte Springs, FL). Tesla has plans to be more engaged here but hasn’t shown anything yet.
Construction and rural focused solutions
One area which players will need to focus on is how to deal with construction areas spaces that affect the road, as well as the off-road situations (e.g. a concrete mixing truck needs to drive on main roads as well as off-road in the construction site), being able to handle such situations, will be vital to the viability of a driverless world.
Additionally, rural areas may not have a fully developed infrastructure, such as paved roads. Driverless vehicles may not actually play a role here, but how does someone who needs to go from point A (urban and driverless) to point B (ill-defined roads and rural) will be something that needs to be figured out).
If we do go fully driverless, handling these hand-off situations will need to be figured out. OEMs may need to provide some mechanism for backup drive by wire technologies. I personally have not seen any efforts by any player in this space.
While there are many players in this space, there seems to be a clear divide between OEMs and technology companies. Technology companies have a huge advantage when it comes to activities around machine learning as well as collecting and processing data. OEMs on the other hand, have an advantage in their ability to manufacture vehicles, and working with municipalities.
Tesla being a bit of both and has provided it with a clear advantage in this space. Google’s efforts in machine learning giving it a strong foothold as well. Uber has a mixed bag of advantages and disadvantages in this space. And OEMs are going to need to partner quickly. It will be interesting to see how all of this plays out.