Lessons Learned When Robots Fail

Pundits are already speculating that the fatal crash by Uber’s autonomous taxi will set the driverless car industry back decades. As the National Transportation Safety Board (NTSB) deconstructs the accident to report on the technological breakdown, it is clear that the redundant systems failed – from the safety driver’s ability to react in time to the vision system’s identifying the risk and activating the brakes. The memory of Elaine Herzberg will live on through our society’s ability to evolve the technology with smart public policy that strives for zero accidents. To accomplish this goal, public safety and unmanned vehicles are eternally intertwined – on the same day that Herzberg was killed, one hundred other vehicular deaths occurred at the hands of human drivers.

In writing the history of robotics one would be hard pressed not to find a correlation between human tragedy and technological advancement. Probably the greatest amount of funding and global interest in automation took place following the triple reactor meltdown of the Fukushima Daiichi Nuclear Power Plant on March 11, 2011. The Fukushima disaster was the result of a megathrust earthquake that triggered a fifty foot tsunami into the Japanese coastal city of 32 million, leaving 19,000 dead in its wake. As a result of the radiation leak, 573 people were killed and more than 160,000 residents evacuated. The toxicity levels have made cleanup efforts nearly impossible for humans, providing windows of opportunities (and a DARPA Challenge) for mechatronic innovators.

It took six years for scientists to develop a robot capable of reaching one of the three reactor cores to locate the burnt uranium fuel rods. The machine, Mini-Manbo (translated as Little Sunfish), is the result of more than a half decade of unmanned system failures. In the shadow of the Little Sunfish is a graveyard of melted metal bots (slideshow below). These technical setbacks have attributed to an exploding multi-billion cleanup budget and public outcry. Within just the first four years of the recovery efforts, Japan allocated close to two trillion yen (or fifteen billion dollars) towards the areas around the Fukushima Daiichi plant. According to the New York Times, Japanese Prime Minister Shinzo Abe declared last year that by the tenth anniversary of the tsunami, workers will begin extracting “the melted fuel from at least one of the three destroyed reactors.” To accomplish this goal, engineers are building more radiation-resistant robots, like Mini-Manbo, at the newly minted $100 million Naraha Remote Technology Development Center. In describing the enormity of the effort, the center’s R&D director, Shinji Kawatsuma, said, “I’ve been a robotic engineer for 30 years, and we’ve never faced anything as hard as this. This is a divine mission for Japan’s robot engineers.”

This slideshow requires JavaScript.

This week the manager of the Fukushima site, Tokyo Electric Power Co. Holdings (Tepco), announced a new approach to surveying the interior of the toxic infrastructure – drones. Tepco contracted with American-based Southwest Research Institute (SwRI) to build a fleet of autonomous unmanned quadcopters to report on the interior conditions of the containment vessels. SwRi is working with the GRASP Lab of the University of Pennsylvania to leverage their leadership in computer vision, simultaneous localization and mapping (SLAM), and swarm mechanics.

Project manager Dr. Monica Garcia stated “the conditions inside the containment at Fukushima Daiichi are quite possibly the most challenging environment that the SwRI-Penn team has had to address. We will be pushing the envelope in terms of the technology.” However, Garcia is optimistic of their chances in aiding the critical work of decontamination and decommission, as recent SwRI’s drone flights on campus were able to survive simulated harsh radioactive environments. In the words of UPenn’s Dean of Engineering and Applied Science, Dr. Vijay Kumar, “Challenges like this are what push research in our field forward. As robots get smaller, faster and smarter, this is exactly the kind of problem we want them to address.”

The catastrophe of Fukushima is a solemn reminder of the wisdom of the late Stephen Hawking: “I believe that the long-term future of the human race must be in space. It will be difficult enough to avoid disaster on planet Earth in the next hundred years, let alone the next thousand, or million. The human race shouldn’t have all its eggs in one basket, or on one planet. Let’s hope we can avoid dropping the basket until we have spread the load.” Hawking’s premonition raises the stakes for roboticists as the survival of not just one Japanese city is at risk but the entire species.

The future of interstellar travel depends on the advancement of automation through trial and error. The International Space Station (ISS) is one of the most expensive projects of the world community with an annual budget of $150 billion, with two-thirds paid for by the United States government. In 2011, NASA began testing humanoids on the ISS with the launch of Robonaut 2 (R2). Adjusting to its new mission, R2 quickly demonstrated its capabilities with flipping switches and passing items to human counterparts. Yet, the long-term goal of having an automated assistant taking over tedious repetitive tasks on the ISS proved more difficult, as mobility became a big obstacle for a robot without lower extremities. In 2014, NASA decided to send the droid legs at sticker price of $14 million (more than five times the cost of the original unit). Unfortunately, the upgrade did not go as planned, and R2 had to be powered down following several failed attempts. Last month, the IEEE reported that the space agency finally boxed the arms, legs, torso and head back to Earth for repair.

In explaining the decision, Robonaut Project Manager Julia Badger said, “The whole point of the ISS is to be able to try different things out. I think [Robonaut] has given us a lot of knowledge on what the requirements for humanoid robots in space will be in the future. We’re bringing it home, repairing it, and in the near future after that, we’re hoping to fly it up there again to proceed with our original goals of advancing new technology.” The failed R2 experiment also yielded several major advancements for medical devices and workplace ergonomics on Earth. According to NASA, the dextrous robot further enabled the evolution of the a new generation of robotic gloves/prosthetics, exoskeletons, and telepresence operations.

In 1896, the first horseless carriage barreled down the the streets of Detroit at a pace of twenty miles per hour described by the local newspaper as “tearing along the street at a lively rate, dodging people and teams.” As accidents and fatalities increased in large numbers, Detroit led the nation by creating the first stop sign in 1915 and eventually the first traffic lights. These safety measures fostered better public confidence between drivers, riders and pedestrians. Since then, roadside tragedies have inspired numerous inventions from brake lights to seat belts to airbags. As autonomous machines are now rolling into our crowded neighborhoods the setbacks ahead are real and sometimes regrettably fatal. Innovators would be wise to listen to the mantra of Apollo 13 Flight Director Gene Kranz, “Failure is not an option.”

This topic and more will be discussed at the next RobotLab event on “The Politics Of Automation,” with Democratic Presidential Candidate Andrew Yang and New York Assemblyman Clyde Vanel on June 13th @ 6pm in NYC – RSVP Today!