AI AND SAFE SELF-DRIVING CARS

FROM:  NATIONAL SCIENCE FOUNDATION
Programming safety into self-driving cars
UMass researchers improve artificial intelligence algorithms for semi-autonomous vehicles
February 2, 2015

For decades, researchers in artificial intelligence, or AI, worked on specialized problems, developing theoretical concepts and workable algorithms for various aspects of the field. Computer vision, planning and reasoning experts all struggled independently in areas that many thought would be easy to solve, but which proved incredibly difficult.

However, in recent years, as the individual aspects of artificial intelligence matured, researchers began bringing the pieces together, leading to amazing displays of high-level intelligence: from IBM's Watson to the recent poker playing champion to the ability of AI to recognize cats on the internet.

These advances were on display this week at the 29th conference of the Association for the Advancement of Artificial Intelligence (AAAI) in Austin, Texas, where interdisciplinary and applied research were prevalent, according to Shlomo Zilberstein, the conference committee chair and co-author on three papers at the conference.

Zilberstein studies the way artificial agents plan their future actions, particularly when working semi-autonomously--that is to say in conjunction with people or other devices.

Examples of semi-autonomous systems include co-robots working with humans in manufacturing, search-and-rescue robots that can be managed by humans working  remotely and "driverless" cars. It is the latter topic that has particularly piqued Zilberstein's interest in recent years.

The marketing campaigns of leading auto manufacturers have presented a vision of the future where the passenger (formerly known as the driver) can check his or her email, chat with friends or even sleep while shuttling between home and the office. Some prototype vehicles included seats that swivel back to create an interior living room, or as in the case of Google's driverless car, a design with no steering wheel or brakes.

Except in rare cases, it's not clear to Zilberstein that this vision for the vehicles of the near future is a realistic one.

"In many areas, there are lots of barriers to full autonomy," Zilberstein said. "These barriers are not only technological, but also relate to legal and ethical issues and economic concerns."

In his talk at the "Blue Sky" session at AAAI, Zilberstein argued that in many areas, including driving, we will go through a long period where humans act as co-pilots or supervisors, passing off responsibility to the vehicle when possible and taking the wheel when the driving gets tricky, before the technology reaches full autonomy (if it ever does).

In such a scenario, the car would need to communicate with drivers to alert them when they need to take over control. In cases where the driver is non-responsive, the car must be able to autonomously make the decision to safely move to the side of the road and stop.

"People are unpredictable. What happens if the person is not doing what they're asked or expected to do, and the car is moving at sixty miles per hour?" Zilberstein asked. "This requires 'fault-tolerant planning.' It's the kind of planning that can handle a certain number of deviations or errors by the person who is asked to execute the plan."

With support from the National Science Foundation (NSF), Zilberstein has been exploring these and other practical questions related to the possibility of artificial agents that act among us.

Zilberstein, a professor of computer science at the University of Massachusetts Amherst, works with human studies experts from academia and industry to help uncover the subtle elements of human behavior that one would need to take into account when preparing a robot to work semi-autonomously. He then translates those ideas into computer programs that let a robot or autonomous vehicle plan its actions--and create a plan B in case of an emergency.

There are a lot of subtle cues that go into safe driving. Take for example a four-way stop. Officially, the first car to the crosswalk goes first, but in actuality, people watch each other to see if and when to make their move.

"There is a slight negotiation going on without talking," Zilberstein explained. "It's communicating by your action such as eye contact, the wave of a hand, or the slight revving of an engine."

In trials, autonomous vehicles often sit paralyzed at such stops, unable to safely read the cues of the other drivers on the road. This "undecidedness" is a big problem for robots. A recent paper by Alan Winfield of Bristol Robotics Laboratory in the UK showed how robots, when faced with a difficult decision, will often process for such a long period of time as to miss the opportunity to act. Zilberstein's systems are designed to remedy this problem.

"With some careful separation of objectives, planning algorithms could address one of the key problems of maintaining 'live state', even when goal reachability relies on timely human interventions," he concluded.

The ability to tailor one's trip based on human-centered factors--like how attentive the driver can be or the driver's desire to avoid highways--is another aspect of semi-autonomous driving that Zilberstein is exploring.

In a paper with Kyle Wray from the University of Massachusetts Amherst and Abdel-Illah Mouaddib from the University of Caen in France, Zilberstein introduced a new model and planning algorithm that allows semi-autonomous systems to make sequential decisions in situations that involve multiple objectives--for example, balancing safety and speed.

Their experiment focused on a semi-autonomous driving scenario where the decision to transfer control depended on the driver's level of fatigue. They showed that using their new algorithm a vehicle was able to favor roads where the vehicle can drive autonomously when the driver is fatigued, thus maximizing driver safety.

"In real life, people often try to optimize several competing objectives," Zilberstein said. "This planning algorithm can do that very quickly when the objectives are prioritized. For example, the highest priority may be to minimize driving time and a lower priority objective may be to minimize driving effort. Ultimately, we want to learn how to balance such competing objectives for each driver based on observed driving patterns."

It's an exciting time for artificial intelligence. The fruits of many decades of labor are finally being deployed in real systems and machine learning is being adopted widely and for different purposes than anyone had ever realized.

"We are beginning to see these kinds of remarkable successes that integrate decades-long research efforts in a variety of AI topics," said Héctor Muñoz-Avila, program director in NSF's Robust Intelligence cluster.

Indeed, over many decades, NSF's Robust Intelligence program has supported foundational research in artificial intelligence that, according to Zilberstein, has given rise to the amazing smart systems that are beginning to transform our world. But the agency has also supported researchers like Zilberstein who ask tough questions about emerging technologies.

"When we talk about autonomy, there are legal issues, technological issues and a lot of open questions," he said. "Personally, I think that NSF has been able to identify these as important questions and has been willing to put money into them. And this gives the U.S. a big advantage."

-- Aaron Dubrow, NSF

THE TRAINING OF A RESEARCH ROBOT

FROM:   NATIONAL SCIENCE FOUNDATION 
A day in the life of Robotina
What might daily life be like for a research robot that's training to work closely with humans?

On the day of the Lego experiment, I roll out of my room early. I scan the lab with my laser, which sits a foot off the floor, and see a landscape of points and planes. My first scan turns up four dense dots, which I deduce to be a table's legs...
Robotina is a sophisticated research robot. Specifically, it's a Willow Garage PR2, designed to work with people.

But around the MIT Computer Science and Artificial Intelligence Laboratory, it is most-often called Robotina.

"We chose a name for every robot in our lab. It's more personal that way," said graduate student Claudia Pérez D'Arpino, who grew up watching the futuristic cartoon The Jetsons. In the Spanish-language version, Rosie, the much-loved household robot, is called Robotina.

Robotina has been in the interactive robotics lab of engineering professor Julie Shah since 2011, where it is one of three main robot platforms Shah's team works with. Robotina is aptly named, as an aim is to give it many of Rosie's capabilities: to interact with humans and perform many types of work.

In her National Science Foundation (NSF)-supported research, Shah and her team study how humans and robots can work together more efficiently. Hers is one of dozens of projects supported by the National Robotics Initiative, a government-wide effort to develop robots that can work alongside humans.

"We focus on how robots can assist people in high-intensity situations, like manufacturing plants, search-and-rescue situations and even space exploration," Shah said.

What Shah and her team are finding in their experiments is that humans often work better and feel more at ease when Robotina is calling the shots--that is, when it's scheduling tasks. In fact, a recent MIT experiment showed that a decision-making robotic helper can make humans significantly more productive.

Part of the reason for this seems to be that people not only trust Robotina's impeccable ability to crunch numbers, they also believe the robot trusts and understands them.

As roboticists develop more sophisticated, human-like robotic assistants, it's easy to anthropomorphize them. Indeed, it's nothing new.

So, what is a day in the life of Robotina like as she struggles to learn social skills?

Give that robot a Coke

I don't just crash into things all the time like some two-year-old human, if that's what you're wondering. My mouth also contains a laser scanner, so I can get a 3-D sense of my surroundings. My eyes are cameras and I can recognize objects...

Robotina has sensors from head to base to help it interact with its environment. With proper programming, its pincher-like hands can do everything from fold towels to fetch Legos (more on that soon).

It could even sip a Coke if it wanted to. Well, not quite. But it could pick up the can without smashing it.

Matthew Gombolay, graduate student and NSF research fellow, once witnessed the act. At the time, he wasn't sure how Robotina would handle the bendable aluminum can.

"I wanted it to pick up a Coke can to see what would happen," Gombolay said. "I thought it'd be really strong and crush the Coke can, but it didn't. It stopped."

That's because Robotina has the ability to gauge how much pressure is just enough to hold or manipulate an object. It can also sense when it is too close to something--or someone--and stop.

Look, I'm 5-feet-and-4.7-inches tall--even taller if I stretch my metal spine--and weigh a lot more than your average human. If I sense something, I stop...

Proximity awareness in robots designed to work around people not only prevents dangerous or awkward robot-human collisions, it builds trust.

"I am definitely someone who likes to test things to failure. I want to know if I can trust it," Gombolay said. "So, I know it's not going to crush a Coke can, and I'm strong enough to crush a Coke can, so I feel safer."

Roboticists who aim to integrate robots into human teams are serious about trying to hard-wire robots to follow the spirit of Isaac Asimov's first Law of Robotics: A robot may not injure a human being.

Luckily, when decision-making robots like Robotina move into factories, they don't have to be ballet dancers. They just have to move well enough to do their jobs without hurting anyone. Perhaps as importantly, the people around them must know that the robots won't hurt them.

Robots love Legos, too

The day of the Lego experiment is eight hours of fetching Legos and making decisions about how to assemble them. The calculations are easy enough, but all that labor makes my right arm stop working. So I switch to my left...

In an exercise last fall that mimicked a manufacturing scenario, the researchers set up an experiment that required robot-human teams to build models out of Legos.

In one trial, Robotina created a schedule to complete the tasks; in the other, a human made the decisions. The goal was to determine whether having an autonomous robot on the team might improve efficiency.

The researchers found that when Robotina organized the tasks, they took less time--both for scheduling and assembly. The humans trusted the robot to make impartial decisions and do what was best for the team.

I have to decide what task needs doing next to complete the Lego structure. The humans text me when they are done with a task or ready to start a new one. I schedule the tasks based on the data. I don't play favorites. When I'm not fetching Legos or thinking, I sit quietly...

"People thought the robot would be unbiased, while a human would be biased based on skills," Gombolay said. "People generally viewed the robot positively as a good teammate."

As it turned out, workers preferred increased productivity over having more control. When it comes to assembling something, "the humans almost always perform better when Robotina makes all the decisions," Shah said.

Predicting the unpredictable

I stand across a table from a human. I sort Legos into cups while the human takes things out of the cups. Humans are incredibly unpredictable, but I do my best to analyze where the human is most likely to move next so that I can accommodate him...

Ideally, in the factories of the future, robots will be able to predict human behavior and movement so well they can easily stay out of the way of their human co-workers.

The goal is to have robots that never even have to use their proximity sensors to avoid collisions. They already know where a human is going and can steer clear.

"Suppose you want a robot to help you out but are uncomfortable when the robot moves in an awkward way. You may be afraid to interact with it, which is highly inefficient," Pérez D'Arpino said. "At the end of the day, you want to make humans comfortable."

To help do so, Pérez D'Arpino is developing a model that will help Robotina guess what a human will do next.

In an experiment where it and a student worked together to sort Lego pieces and build models, Robotina was able to guess in only 400 milliseconds where the human would go next based on the person's body position.

The angle of the arm, elbow, wrist... they all help me determine in what direction the hand will go. I am limited only by the rate at which sensors and processors can collect and analyze data, which means I can predict where a person will move in about the average time a human eye blinks...

Once Robotina knew where the person would reach, it reached for a different spot. The result was a more natural, more fluid collaboration.

Putting Robotinas to work

I ask myself the same question you do: Am I reaching my full potential?

While Robotina's days now involve seemingly endless cups of Legos, its successes in the MIT lab will eventually enable it to become a more well-rounded robot. The experiments also demonstrate humans' willingness to embrace robots in the right roles.

To make them the superb, cooperative assistants envisioned by the National Robotics Initiative--to give people a better quality of life and benefit society and the economy--could require that some robots be nearly as dynamic and versatile as humans.

"An old-school way of thinking is to make a robot for each task, like the Roomba," Gombolay said. "But unless we make an advanced, general-purpose robot, we won't be able to fully realize their full potential."

To have the ideal Robotina--the Jetsons' Robotina--in our home or workplace means a lot more training days for humans and robots alike. With the help of NSF funding, progress is being made.

"We're at a really exciting time," Gombolay said.

What would I say if I could talk? Probably that I'd really like to watch that Transformers movie.

-- Sarah Bates,
Investigators
Julie Shah
Related Institutions/Organizations
Massachusetts Institute of Technology
Association for the Advancement of Artificial Intelligence