It’s Almost Certain That Google’s Self-Driving Cars Will Pivot (or Perish)
For years now, tech giants like Apple, Google, Tesla, and Uber have been fixated on disrupting, if not destroying, America’s auto making industry. Since 2009, Google has led the way with its drive towards fully automated vehicles. Lately, however, the crowding of the competition has made things more problematic.
As it turns out, producing a bad car is easy, but designing and commercializing a fully autonomous one is really freakin’ hard.
Forty Years And Counting
Forget about Google’s ambition to create a fully autonomous car for a moment. When we look at history, we discover that this experiment was run before with 40 years worth of examples.
Despite some nearly ideal closed environments and adequate technology, we still have very little fully autonomous transportation available to us.
Airlines still rely on pilots in the cockpit even though much of the flying is handled by computers.
Ships still have crews even though it’s theoretically possible for them to sail from port to port by themselves.
The NASA Apollo program, which put US astronauts on the moon six different times, was intended to be automated with astronauts as nothing more than passengers. In the end, the astronauts handled many critical functions, including the moon landings.
At all times, humans were the connective tissue that held the system together, constantly making small corrections and picking up mistakes.
When you consider this historical data and couple it with the fact that Google’s self-driving cars can only function on roads that are mapped and perfectly updated by not one but two (!) self-driving mapping cars, you’re quite likely to face a roll-out problem.
This utopian moonshot project then becomes more brittle and a less functional solution to human-centered automation, especially if it comes at a time where almost every other major player in the auto making industry is not following that approach towards self-driving vehicles.
The driverless cars of the future don’t have to replace the drivers; they just need to be smart enough at dealing with bad driving behavior.
Moonshot No More
Ever since the early days of their moonshot program, Google has very much remained in experimental mode. Its fleet of 60 self-driving cars has covered more than two million miles around four U.S. cities — Mountain View, CA; Austin, TX; Kirkland, WA; and Metro Phoenix, AZ.
The Google team has been releasing monthly reports of cumulative miles driven autonomously and manually, as well as any accident reports in those very localized and controlled testing environments.
Their purpose is clear; it’s safer to make the transition towards completely autonomous driving by gradually capturing the world using cameras, radars, and laser sensors rather than commercializing an early version of that product.
Meanwhile, another big contender in this space, Tesla Motors, is already selling cars with some rudimentary self-driving function to the tune of 25,000 per quarter, while incrementally introducing more advanced autonomous features. Also, 70,000 Teslas have already covered over 222 million miles using Autopilot mode — the company’s early bid to create fully self-driving vehicles.
Back in 2009, it was clever of Google to invest in that area and technology before it became popular. Come 2016, it’s already a crowded market with significant competition from a host of new players, including the world’s traditional automakers. Ford, BMW, Volvo, Comma.ai, Didi Chuxing, and others have made significant investments to develop their own self-driving systems through acquisitions and research lab expansions. When this happens, potential investments cease to be opportunities and instead become efficient ways of destroying wealth.
If Google plans to succeed in the self-driving space, then it should be willing to focus on the underlying need that customers want fulfilled. That means, foregoing the testing environment and putting out a testable autonomous self-driving car in the hands of normal users.
Many ambitious projects fail badly because they rely on a Big Bang delivery strategy i.e. making sure that the project is 100 percent done and delivering it at the end. Yet, up until now, Google’s prototypes have all been operated on by Google engineers. That means that over the past seven years, a lot of time has passed without any user actually testing the product in an uncontrolled environment. These cars are then more likely to be riddled with design flaws based on incorrect assumptions of what people need (or want to hear, like this poor fella who test drove a Google self-driving car and almost ended up in jail.)
The absence of constant and frequent feedback loops and a heavy internal focus on iterating for a perfect scope has made Google’s self-driving cars a very risky project.
“No matter how much up front analysis you do, you’re still surprised when you put the first release in the hands of a real user.” — Henrik Kniberg, LEGO
Rather than anticipating the different driving conditions that riders would face, Google could start collecting more diverse data by putting a beta version of its car on the market. The company can then gauge reports on the usability of the vehicle with real users — including the design of the car, speed limits, or voice recognition settings.
Think of it this way. If software and hardware companies in the 1950’s had held off releasing computers until they’d successfully passed the Turing Test, where would we be today… Exactly.
The sophistication of the hardware and software required to design self-driving cars shouldn’t prevent Google from releasing it to real users. An early release is probably the best way to gradually build trust with an activity that is so ubiquitous to daily life. Establishing this trust is crucial to ensuring people make the decision to fully embrace a new way of using automobiles.
Which is why an incremental data-driven approach involving steady improvements to both a product’s hardware and software, combined with increasing the amount of real world experience, is key towards creating a robust, reliable, customer-lovable product.
And it’s not just Google. Any organization that’s in the midst of a multi-year project development can reap the benefits of quickly getting feedback around an early usable product and avoid the fate that plagued LEGO Universe, New Coke, and Healthcare.gov.
Future project leaders or executives can learn from the unravelling of Google’s approach to driverless cars by paying attention to the following lessons:
User testing — most of the innovation for driverless cars was based on assumptions of what Google thought the customer wants, not on actual feedback from product users. Rather than obsessing with novelty and doing new stuff, you can try to better understand what the customer wants and who they are instead. Focus on what users of the product are saying and how they’re trying to contribute ideas towards making the product better.
Team structure — good executives and project leaders do not disregard the fundamentals of how to run teams with innovative projects. It won’t matter how ambitious your project is, you will need to have a clearly delineated purpose and execution strategy with all members that are working on it. Getting the teaming fundamentals right — which means creating Mission Based Teams, Agile Squads or, a Team of Teams — is key to reviving and governing a stalling innovation project. By bounding innovation with an improved team structure, you’re far likely to get a better payoff than by innovating with no boundaries.
In all likelihood, I don’t think that the fully driverless model will be successful for technical and social reasons. It is unneeded, and it’s historically invalidated and outdone by a more perfect version of partial autonomy where machines don’t replace but team up with humans.
Will a driverless car dare to nudge into traffic? Or take the decision to dodge quickly between a gap in cyclists? Or are we likely to see more vehicles terrified of moving for fear of litigation?
Like the automated transportation technologies that have come before, the best driverless cars of the future will be those with a blend of human awareness and technological affordances.
What approach will you endorse?