Self-driving cars. You’ve been hearing about them for years. While companies like Waymo, Cruise, and Zoox have spent over $16B to deploy vehicles, most still haven’t deployed driverlessly on public roads. It makes you think that teaching a car to drive better than a human has got to be expensive as hell.
Just this year, Waymo (alongside Cruise considered the leader in autonomous driving) had an incident that exemplifies the issues with black box machine learning (ML) systems. A Waymo robotaxi erratically stopped and started in the middle of traffic due to widely spaced traffic cones.1
This sort of scenario is part of the nightmarish “long-tail” that has proven extraordinarily difficult for 100% autonomous systems to handle. ML systems work extremely well on trained data but can have unpredictable behaviors in totally new scenarios. It’s unclear when these fully autonomous systems will have trained on enough edge cases that they will stop having these issues, but progress to date has been slow and gradual. For example, Waymo has only deployed 300 fully driverless robotaxis for public use.
Not only do these autonomous vehicles have to safely avoid collisions on the road, they also have to make passengers comfortable while being driven. That means accounting for a whole host of human factors, like avoiding hard braking or odd swerving, in addition to general safety.
Costs are also a big challenge in autonomous vehicle designs. If you’ve driven around San Francisco recently, you may have seen autonomous vehicles -- Chrysler Pacificas, Jaguar iPaces, or Chevy Bolts -- that have large, expensive LiDARs spinning on them. Waymo recently said their system costs no more than a “moderately equipped Mercedes Benz S-class.” That’s ~$130K for those of you who don’t regularly shop for Mercedes Benz vehicles. There is no way this competes with the cost of a typical Uber driver behind the wheel of a used 2006 Honda Civic.
One main expense has been LiDAR. A driverless vehicle without LiDAR is considered fundamentally unsafe, making LiDAR a “must-have”. LiDAR’s main purpose is collision avoidance. It’s a ranging technology that immediately tells you how far away an object is and its rough shape. But it comes with significant downsides -- expense, range limitation, and increased processing power. Herculean efforts in the past decade to design and build production-ready LiDAR for scale have reduced the price from $75K to sub-$10K for the hardware alone. Maintaining an operational LiDAR unit -- something that has to spin up to 10 times per second during the entire operating life of a vehicle -- requires a lot of work by technicians. A typical electric vehicle can last at least two hundred thousand miles without significant powertrain maintenance. Most LiDAR units are likely an order of magnitude less robust.
On top of all that, the past few years have taught us that teleoperation is still required for autonomous vehicle deployment. This introduces its own issues. Teleoperation is high latency, high workload, and extremely expensive. A remote operator will have a full vehicle control setup (steering wheel, brake pedal, etc) and ingest more visual input than a typical driver, resulting in a high workload. Because every single action goes through the network, there is additional latency between an input from a remote operator and the response on the vehicle.
In the backdrop of all of this is the climate crisis and increasingly gridlocked cities. Autonomous vehicles, as currently being tested, all utilize vehicles like cars or SUVs. A typical SUV in the U.S. weighs 4,000 lbs and gets 25 mpg (or 90 MPGe if electric). A passenger vehicle weighs at least 3,000 lbs and gets 30 mpg (or 100 MPGe if electric). Even the Nuro R2 weighs 2,500 lbs. In driverless cars today, most of the energy is used to transport the car rather than what’s being moved by the car because the vast majority of the weight is the vehicle itself.
Faction to the rescue! Faction is reducing the required R&D to put a driverless vehicle on the road by 4 orders of magnitude 🤯 and reducing the hardware cost by 80%, all while rightsizing the vehicle size and weight. To put this in perspective, four orders of magnitude is the difference between the cost of a used Boeing 727 and this (pretty sweet!) Star Wars Lego set. Put another way, instead of having to hire enough software and hardware engineers to fill the Empire State building, Faction can launch a driverless vehicle with just three people building, testing, and iterating in a garage.
Why is this possible now? Partly because there’s been a rush to build out the infrastructure layer that every self-driving company needs, like processing hardware, data ingestion, sensor fusion, and simulation. Long gone are the days of rolling your own software stack. Platforms like NVIDIA DRIVE are enabling full stack driverless capabilities and speeding up product development in the process.
Faction limits ML systems to highly deterministic functions like lane identification. This limits the complexity and provides clearly defined operational design domains in which the driverless system can be trusted to work. It’s also simpler to validate highly deterministic systems. Faction can also build statistical models around the chance of failure in these systems, mimicking the approach other safety-conscious industries have towards safety critical systems.
Faction is also able to deploy quickly because they don’t need expensive vehicles.
The company successfully lowered vehicle costs by switching to a smaller format vehicle -- the three-wheeled motorcycle. These provide the stability and performance of a typical passenger car, with significantly lower costs -- the average entry two-wheeled motorcycle price in the U.S. is $4-6K new! This means cheaper purchase prices, lower maintenance costs, and better fuel economy.
The switch to motorcycles also enables faster iteration on the hardware stack. Building a driverless vehicle requires supplier support to get safety hooks into every major system involved in driving. For a typical passenger vehicle, the automotive supply chain is geared toward mass volume with minimum order quantities in the tens of thousands and no incentive to work on small volume programs. By contrast, powersports suppliers will sell in low volumes and are willing to use engineering resources to design driverless-ready parts for the chance at a market that can add 10K vehicle sales per year.
Faction chooses to use robust, production-ready radar, cameras, and ultrasonic sensors that can be found on any car with collision avoidance, cruise control, and lane keeping functionalities. It uses these sensors in a layered safety system that can cross-check to confirm each sensor is functioning correctly and provide a well-defined operational design domain to operate in. This eliminates the need for costly, research-grade LiDAR and camera systems.
In order to speed deployment, Faction doesn’t plan on carrying people initially. Instead, they’ll use their driverless system to deliver cargo. For rideshare, Faction will use the driverless system to move the vehicles to a user, who will then drive themselves. This eliminates human factors that autonomous driving systems must design around. Cargo doesn’t care if you have to hit the brakes in order to avoid a collision, but people do.
Faction will deploy their driverless vehicles with integrated teleoperations. Most people don’t realize that completely removing teleoperation doesn’t actually result in significant cost savings. Starsky Robotics reduced teleoperation time on highways to 1% and paid a remote teleoperator a salary of $60,000 to yield an average cost of teleoperation to $600 per truck per year. Even though Faction will operate in cities and therefore spend a higher amount of time in teleoperation than trucks on highways, there’s still more opportunity to reduce the vehicle bill of material cost than there is in funding research projects to attain 100% autonomous driving.
Faction’s take on teleoperation actually transforms autonomy from a product to an optimization problem. Instead of requiring its autonomous system to handle 100% of edge cases that are currently holding up deployment in full autonomy projects, Faction can continually increase the operational design domain of the driverless system to reduce teleoperation.
Switching to three-wheeled motorcycles also right sizes transportation for the typical trip. An electric, three-wheeled motorcycle will typically only weigh 1,000 lbs and get way more than 150 MPGe. That’s 75% less mass than a typical SUV and an over 60% increase in fuel efficiency. The motorcycle is also physically shorter and narrower, translating to less congestion on the road and in parking lots.
Faction’s initial strategy is to deploy in a market they are uniquely positioned to win -- micrologistics. They’re targeting delivery payloads less than 400 lbs between two defined points where the route is constantly run, such as a hub and spoke delivery model for businesses like central kitchens or bakeries. These deliveries are often made in the early morning when it is difficult to find a driver, adding to customer frustration. This strategy is also applicable for businesses requiring immediate shipment of a product to keep serving customers, like a coffee shop that’s run out of an ingredient mid-day or an auto body shop requiring a specific part to proceed with a repair. In all of these micrologistics use cases, Faction is already cheaper than any option with a paid driver -- even before scaling manufacturing.
While in the near term Faction will design, build, and deploy a small fleet of driverless vehicles, Faction’s long term vision is to provide the core technology for the industry to go driverless. They’ll provide reference designs for how to build the driverless hardware with all the safety hooks required to run Faction’s core system, DriveLink. This will enable any supplier or OEM to build driverless vehicles with DriveLink. Those vehicles can then be sold to transportation and logistics companies who specialize in operating fleets for transportation networks to move people and goods. Faction-enabled vehicles will be used in B2B services where APIs are used to dispatch vehicles.
In this vision, Faction will have achieved what few other self-driving tech companies have achieved -- brought driverless vehicles to public roads for all to use! They’ll accomplish this with healthy margins (even at a small scale), on a tighter timeline, and with downsized vehicle size and weight to directly lower carbon emissions to move people and goods around.
There is no one more equipped to do this than the founder and CEO of Faction, Ain McKendrick. He previously led engineering at Starsky Robotics where they built and ran the first driverless (no safety driver in the cab!) semi-truck on public roads legally in the U.S. While at Starsky, he oversaw development of teleoperation systems. Additionally, Ain previously founded 3 companies and was the VP of Engineering at Cygn, a developer of self-driving vehicles.
At Fifty Years, our sweet spot is supporting founders at the earliest stages who are building deep tech companies that can generate huge financial outcomes and create massive positive impact.
Deep tech: Ain is creating the foundational operating system at the intersection of hardware, software, and human operations that will enable scaling safe, cost-effective electric driverless systems everywhere.
$1B yearly revenue potential: The movement of goods and people is one of the largest markets on the planet. The courier and local delivery market in the U.S. alone is $120B. By providing the first driverless system capable of delivering goods, Faction will be able to offer a lower cost per mile than any competitor with a paid driver in micrologistics and rideshare services, all with a healthy margin.
Massive positive societal impact: Faction will deliver on the driverless future, now. There are over 30,000 fatalities a year on U.S. roads alone -- most of which are due to human error and would be preventable with driverless systems. Faction will be able to decrease the number of traffic-related fatalities. They’ll help speed up the transition to electric vehicles. They’re also increasing energy efficiency and decreasing congestion by downsizing vehicle size.
Inspired by their vision of deploying and scaling electric vehicles and driverless technology, Fifty Years is excited to lead Faction’s seed round alongside Trucks VC. At Fifty Years, helping great scientists and engineers become great entrepreneurs is our jam, and we’re looking forward to helping Ain and the Faction team deploy and scale electric driverless vehicles faster than anyone thought possible!
Taken from this CNN article: “First, the Waymo vehicle paused at a stop sign rather than turning onto a street lined with cones. The car then completed the turn, but soon stopped in the road, blocking part of a lane of traffic. Following a four-minute stop, it backed up slightly, further blocking a traffic lane. Human motorists had to cross a double yellow line to go around the Waymo vehicle. Some honked. A construction crew removed a cone in the Waymo vehicle's path, but the car remained stopped. A few minutes later, the Waymo car pulled away, surprising a Waymo worker who was explaining to the van's passenger, Joel Johnson, through the car's audio system that roadside assistance was on its way. Further down the road, the Waymo van halted again, amid yet more cones. It was then that a Waymo roadside assistance vehicle arrived. Then, as the human driver approached, the Waymo car drove away again, but only a short distance.”