Automation’s “Goldilocks” Challenge: Why Fox Robotics Is Just Right
People tend to overestimate how disruptive a certain technology will be in the short term, but underestimate its influence in the long term. In venture capital investing, this manifests itself as a flood of capital into an exciting new technology that is simply too immature to commercialize, or a market that takes too long to develop. With that comes a whole crop of startups that go out of business.
This has certainly been the case with autonomous vehicles (AVs), as illustrated by the recent acquisition of Zoox by Amazon for about $1 billion—which sounds like a nice exit except that they raised around $1 billion in equity and debt to get there. And that’s one of the better outcomes!
The meltdowns are coming for autonomous cars, trucking, robo-taxis, and last-mile delivery services. There are way too many companies chasing the autonomous future of transportation, and here’s the main hitch: it doesn’t really work yet.
Driving is something people can do pretty well, but it is an “unstructured” task. Driving at “Level 5”—which requires no safety driver and the ability to drive any arbitrary route—is out of reach for the most sophisticated AI autopilots today. Will we get there? Yes. But it will be years from now, and the path will be evolutionary.
AVs that navigate “semi-structured” environments and operate at “L4” or lower will be first, operating now as shuttles in gated communities, food delivery within the confines of a college campus, and taxis that follow prescribed routes (all with safety drivers, of course). Long haul trucking is a promising first candidate too, but even that is hard for AI today under real-world conditions. We are getting there, and the work that is being done is truly amazing, but the stakes are incredibly high. An autonomous vehicle has to be as safe as a human driver—preferably safer—and cheaper. It’s a high standard to achieve.
The history of innovation is littered with startups who were too early or too late.
At Menlo Ventures, we frame this “Goldilocks problem” of getting the timing just right as a question: “When is there a convergence of all the necessary factors to make this big problem tractable?” Ideally, you want to solve something that is hard enough so that not everyone can do it, but it’s not TOO hard.
After poking around the AV market for a while, we broadened our scope to include autonomy and robotics broadly. We’ve witnessed the convergence of multiple technologies: cheap and massive computing power, adequate machine vision, low-cost sensors, cameras and lidars, powerful machine learning algorithms and AI. In just the last few years, robots have advanced from doing highly structured, repetitive tasks (think of an automotive manufacturing bot) to taking on much more complex “semi-structured” tasks.
The impact of this step-function improvement in utility is enormous, and one of the first places to see that impact is warehouse automation. E-commerce is causing major changes; a far greater number of smaller orders that have to be picked and packed instead of shipped out to retail stores is causing major strains on backend order fulfillment in addition to labor shortages.
6 River Systems was our first investment in this area and was recently acquired by Shopify, but there is far more to be done. RightHand Robotics is automating the picking of items with a self-learning and highly versatile gripper. And our latest investment, Fox Robotics, is making autonomous forklifts that load and unload trucks—a typical large distribution center may have up to 150 or so of these, and unload hundreds of trucks per day.
These are not easy problems for today’s autonomous bots. Even when the environment is controlled—the object you have to pick up is in a bin with a well-defined boundary, the pallet you need to unload is wooden and has spaces for the forks on the lift to fit into—it can get messy. Wooden pallets break, loads shift, lighting changes, someone bumps you. Forklifts and picking arms still need human intervention when they get confused. And that is fine when you are in a warehouse and the robot can replace 90 percent of what a person can do. It is not fine when you are a vehicle on the road.
Where else are robots ready to produce big ROI? We’ve looked for other problems that fit the “Goldilocks” model. Two industries where we see potential are construction and agriculture. Certain assembly tasks and piece-finishing like welding and painting are good candidates. At some point, robots could be a huge aid to our aging population and dramatically reduce the cost of home healthcare, but like L5 autonomous vehicles, this is still a futuristic vision and not ready for VC investment. I suspect the robotics category is going to be a good one for VCs and entrepreneurs for some time to come. The trajectory of technological advancement makes me optimistic that machines will, in fact, be capable of taking on totally unstructured tasks and likely perform them better than we do, but along the way, we need to be careful not to invest ahead of reality. The team at Fox are some of the best