Observations & Opportunities: Autonomous Vehicles (AVs) Powered by AI—What Could Possibly Go Wrong?

This time, the AV discussion continues with some general and specific Artificial Intelligence (AI also known as Machine Intelligence, MI) downsides, challenges and tips. Autonomy, safety and efficiency are obvious AI goals for AVs. AI is a tool ranging from simple to complex. If AI is implemented with a thorough understanding of the specific goals to accomplish, it may work very well and exceed expectations. If not, watch the headlines.

AI is the fast-evolving foundation for many new technology efforts now including self-driving cars, a.k.a., AVs. There are as many different approaches to AI for AVs as there are companies working on them and the AVs themselves. Automakers, with their own systems, or those that are evaluating the alternatives from Waymo by Google (Alphabet), Apple, or others, will cull the field at some point. Some systems will be too expensive and others may rely upon unproven technologies that will turn off auto manufacturers. Some standardization is necessary too for regulation and insurance purposes. AV appearance will be a factor since some of the roof-mounted sensor arrays are just plain ugly and need cloaking devices. Some of the non-aerodynamic sensor arrays on roofs would increase noise and decrease mileage.

Waymo’s Chrysler Pacifica Hybrid minivan AV testbed

As Waymo points out, the world had 1.25 million deaths worldwide due to vehicle crashes in 2014. There were 32,675 deaths in the U.S. alone due to vehicle crashes in 2014. And there was a 6% increase in traffic fatalities in 2016, reaching the highest point in nearly a decade. The primary cause: “There’s a clear theme to the vast majority of these incidents: human error and inattention.” Safety is a primary Waymo goal.

Apple’s AI for AV efforts are usually somewhat secret. However, they recently posted their Apple Machine Learning Journal online. It opens with “Welcome to the Apple Machine Learning Journal. Here, you can read posts written by Apple engineers about their work using machine learning technologies to help build innovative products for millions of people around the world. If you’re a machine learning researcher or student, an engineer or developer, we’d love to hear your questions and feedback.” It provides some insights on what they are doing.

How much AI is used in products or services including AVs varies enormously but it is the elephant in the room for all “smart” products whether it be your smartphone, tablet, car or thermostat. Like them or not, AI-enabled products and services can turn up anywhere and there will be some inevitable marketing hyperbole about new “AI” products. AI-enabled things may be smarter than they were but they still lack common sense and often adequate security.

A recent publicized fender-bender accident involving a newly deployed AV bus suggests that an AV should have enough latitude to back up, if no one is endangered, when an obvious threat is posed by another vehicle backing up towards it. Of course, some human drivers would have not taken evasive action either so blaming AI for lack of foresight or adequate programming is really blaming humans too.

AI is, and likely will be for some time, a blend of hardwired programming for following rules of the road as well as having a ML (machine learning) capability to avoid repeating mistakes and learning how to perform better. There certainly are differing approaches for an AV’s operating system (OS) depending, in part, on whether it was conceived by an old-line auto company or an new entry such as Apple or Waymo (Alphabet/Google).

How Extensive Is AV and AI?

One insightful recent report proclaimed, “There are basically three big questions about artificial intelligence and its impact on the economy: What can it do? Where is it headed? And how fast will it spread?

“Three new reports combine to suggest these answers: It can probably do less right now than you think. But it will eventually do more than you probably think, in more places than you probably think, and will probably evolve faster than powerful technologies have in the past.” This suggests that robust global AI efforts are not only underway but gaining momentum.

As a disclaimer, most of the companies describing AI advancements now do so relative to their own efforts. A dispassionate comprehensive overview of AI can be found online by visiting the Artificial Intelligence Index and reading their AI Index 2017 Annual Report. For those considering or upgrading AI in their products or services, the report is very revealing and allows one to avoid some common pitfalls.

AI, AV and EV Futures: Defined by the Giants

In AI, the really large technology leaders and their competitors are battling it out in the AI marketplace. It is a battle where substantial financial resources and engineering talent help carve out significant market shares. So, are these large corporations too influential?

“We are experiencing a time, where five companies are holding most of the economical (and even political) power in the world: Facebook, Alphabet, Amazon, Apple and Microsoft” surfaced in a recent blog. Alphabet (Google and Waymo), Tesla (SpaceX and the Hyperloop Pod transport-in-a-vacuum system), Uber and Apple are certainly active in AI for AVs.

Some AI efforts may require more than what those companies alone can accomplish alone or with a few partners. If one has any doubt that AI, or AI in AVs, is real, just check out the news and recent technical papers presented by these companies and their automobile manufacturing competitors around the world. Of course, the automobile manufacturers are very active which is no surprise. Don’t overlook the universities and R&D centers are doing AI and AV research.

Many of the biggest companies involved with AI projects had modest efforts until the past few years, the exception being Waymo. AI and its impact on AVs, EVs (Electric Vehicles) and other products and services were modest but with some very promising results. Some of the deep-pocket companies were busy with projects that were not yet public knowledge. However, with AVs, General Motors just announced that they are “all in” with the AV and EV businesses going forward.

G.M. produced their first round of AV self-driving Chevrolet Bolt cars at their GM Orion Assembly plant in Orion Township, Michigan. G.M.’s president, Daniel Ammann, told journalists that the cars would be ready for consumer applications in “quarters, not years.”

G.M. also plans to build a driverless ride-hailing fleet by 2019 that may eventually become part of the automaker’s core business. [G.M. the taxi company?] Executives told investors that by 2025, AV cost reductions and increased consumer purchases should combine and drive prices down to less than $1/mile, or about a third of current ride-hailing prices.

G.M.’s Chevrolet Bolt EV is the first U.S.-made, mass-market, fully electric car

“We’re aiming for a future of zero crashes, zero emissions and zero congestion,” G.M. CEO Mary Barra said.

Aside from the AV future, G.M. executives said the company sees a clear path to profitability through a wide array of EV electric cars, which so far have yet to take hold with consumers due to a combination of cost and range anxiety but the latter may be reduced as battery technology is improving.

Innovations Welcome

The current breakthroughs in commercializing AI, AVs, EVs and related technologies will provide opportunities for those with innovative ideas. With some companies now offering do-it-yourself (DIY) ML (Machine Learning) apps, it may get much more interesting quickly.

The AI genie is out of the bottle. Governments, corporations, schools and individuals certainly have increasing access to AI technologies. One major question is whether or not they will make informed decisions regarding selections, implementations and usage. The general public does not have an in-depth knowledge of what is being planned or secretly deployed whether it be for the greater good or for nefarious purposes. However, the better that AI is understood by everyone, the more likely consensus positive outcomes will be implemented.

It Is AV Rocket Science

Yes, space launch rocket systems can be AVs too. Elon Musk, of Tesla automotive fame, sees his impressive SpaceX space launch venture on a roll. Their satellite launches in 2017, to date, are impressive and are leading the way to more affordable launches among global space launch providers. It’s nice to see a private company succeed in an area long dominated by government or existing aerospace companies. Do reusable SpaceX rockets use AI? Yes. In fact, AI was the tool that allowed analysis of the key engineering challenges that make landing huge rockets possible.

No entity had seriously studied re-entry and landing the first stages of the rockets until recently. Companies and governments had just figured disposable rockets were the cost of doing business since everyone used them. Even the now defunct U.S. Space Shuttle’s original reusability goals faded over the years for numerous reasons.

A Falcon 9 first stage successfully landed on SpaceX’s autonomous spaceport drone ship in the Atlantic Ocean. This first stage was reflown in March 2017, the first-ever reflight of an orbital-class rocket stage.

Musk is certainly not alone in the private space launch business, other tech billionaires including Paul Allen [co-founded Microsoft with Bill Gates, and founder of the Allen Institute for Artificial Intelligence and Stratolaunch Systems, a space transportation venture using an air launch to orbit system] and Jeff Bezos (Amazon and Blue Origin spacecraft founder) may be enabling the new age in space launch efficiency and affordability. These new “kids” are so good that even the old Russian standby Soyuz team is scrambling to rework their business to compete.

The Soyuz spacecraft was designed for the Soviet space program by the Korolev Design Bureau (now RKK Energia) in the 1960s. The Soyuz was originally part of the unrealized Soviet manned lunar manned landing program. The Soyuz capsule launches on a Soyuz rocket, a frequently used and very reliable launch vehicle. The Soyuz rocket is based on the early Vostok launcher, which in turn was based on the 8K74 (R-7A Semyorka) Soviet-era ICBM.

According to one recent report, “This year [2017] has seen a number of firsts for the [SpaceX]—first reflight of a Falcon 9 booster, first reuse of a Dragon cargo spacecraft, first national security payload, and a remarkable dozen landings. But probably the biggest achievement has been finally delivering on the promise of a high flight rate.”

Yes, these entrepreneurs are using AI to fine-tune their space transportation operations as well as their other commercial enterprises. Yes, AI works.

Using AI for AVs & Everything Else

Although science and engineering innovations are built upon the accomplishments of the past, fresh eyes from younger visionaries and the curious may shake up industry and business going forward. When the the young ask “Why?”, or see something that looks overly complicated, fresh insights may lead to greater efficiency and productivity.

Musk is both a user and promoter of AI. “Since its founding by Elon Musk and others nearly two years ago, nonprofit research lab OpenAI has published dozens of research papers. One posted online … is very different: Its lead author is still in high school.

“The wunderkind is Kevin Frans, a senior currently working on his college applications. He trained his first neural net—the kind of system that tech giants use to recognize your voice or face—two years ago, at the age of 15. Inspired by reports of software mastering Atari games and the board game Go, he has since been reading research papers and building pieces of what they described. “I like how you can get computers to do things that previously you would think were impossible.”

Frans was working on a tricky problem that was holding back robots and other AI ML systems, i..e., “How can machines tap what they’ve previously learned to solve new problems?”

Humans do this instinctively but machine-learning (ML) software typically repeats its lengthy training process for every new problem—even when there are common elements.

Frans’s paper, with four others affiliated with the University of California Berkeley, reports progress on this problem. “If it could get solved it could be a really big deal for robotics but also other elements of AI,” Frans says. He developed an algorithm that helped virtual legged robots learn which limb movements could be applied to multiple tasks, such as walking and crawling. In tests, it helped virtual robots with two and four legs adapt to new tasks, including navigating mazes, more quickly. Fresh ideas help.

Next time: AI-driven AV technology will certainly reshape the global automobile manufacturing industry and its suppliers. Manufacturing innovations are likely that will reshape the supply chains and manufacturing landscape. A look at some possibilities and additional concerns from both automakers and tech entrepreneurs will provide additional AI and AV insights.