In a world full of self-driving cars, flying drones, and other robots, daily interactions with artificial intelligence will have a profound effect on how we live our lives. Elemental video scientist Boonsri Dickinson visited Carnegie Mellon robotics pioneer Maneula Veloso to talk about the science behind her robotic creations and the many years she has spent bringing autonomous robots to life. For video see this link and embedded below:
Why do we create certain expectations when interacting with technology? Who or what informs these expectations? While humans carry cell phones and computers, we would never expect these machines to be able to serve us a cup of coffee. However, when designing robots to perform tasks, learn, and roam, a great deal of thought must be put into the environment, and the expectations that form along with it.
What qualities make a robot, a robot? What human qualities would researchers like robots to have? Getting to the heart of this very question is Manuela Veloso, a leading artificial intelligence professor at Carnegie Mellon University, who has robots that depend on humans for help as they navigate their way around her lab.
Manuela is interested in understanding robots and their interactions with humans. What is an intelligent robot? How can robots not only co-exist with, but also help humans to excel?
Designing a robot to perceive the world is key to making an intelligent machine that can interact with other robots, interface with information on the internet, and talk to humans that occupy the same space. Perception is a very human quality, one that occurs through our five senses. Robots do not share these senses, so technology must be used to create cognition abilities and an artificial body to carry out a set of programmed actions.
Manuela said that “I feel very strongly about my vision about things that move by themselves. [You can tell the robot to] please go there and the thing will go. Can you see if I left things in my office? Can you escort this person?”
Asking For Help
While the CoBot robot isn’t as cute as science fiction prototypes such as WALL-E, the rigid, lanky robot does have one very charming characteristic: It is humble.
A CoBot robot is designed to be an autonomous intelligent mobile robot system. CoBot robots know they do not have arms and must ask humans for help.
For example, a CoBot robot would greet a visitor to the lab and then offer the guest a cup of coffee. To perform tasks such as getting coffee, the CoBot must first ask humans to place a cup of coffee on its base. The robot would then continue to map out its environment, figuring out where to deliver the coffee.
“A lot of these AIs are about creating fully autonomous robots, but when putting robots into human environments, they’re [environments] very complex. One day about four years ago I realized the issue that robots would have limitations,” Manuela said. By embracing the limitations, she began thinking about autonomy differently. This allowed her to code the robots so that they did not have to be supervised when roaming around freely in her lab.
Manuela’s robots are coded in C++, Python, and Java; they can also use localization and navigation algorithms to move around. Their eyes are nothing more than a Kinect depth-camera, which has access to Wi-Fi and remote sensing technology (LIDAR). Manuela’s robot designs promote efficacy by making the robots ask humans to assist them.
When Manuela began coding robots to play soccer in 1997, she admits that her team didn’t have algorithms to allow the robots to process vision in real-time. As a result, the ball would go one way and the robot would go the other. The game must have been sad to watch, as the robots hardly moved.
It was during 1997 that Manuela helped found RoboCup, an international robotics competition, to promote artificial intelligence and robotics research. It is the stated goal of the project that by 2050, a team of soccer robots will compete with and beat the defending FIFA World Cup champion.
Scientific research is iterative, and it can take years of trial-and-error to design a better system. Now, the robots use algorithms to figure out their position and to score a goal; they can even emote when they are happy by raising their arms.
Still, to solve the problem of image processing, Manuela developed a technique called depth image analysis, which enables a robot to look at an object and detect the ground and walls behind it. The images are processed in real-time, so the environment can be mapped out as the a robot navigates.
“So if you think about an autonomous robot, this is it, they have to have capabilities of doing perception, of doing cognition of actuating, of coordinating with each other, coordinating with people and accessing the web,” Manuela said.
Learning To Live With Robots
Google set up a robotics group and acquired eight robotics companies last year, according to an article in the WSJ. As MIT Center for Digital Business director Erik Brynjolfsson points out in his book, The Second Machine Age, self-driving cars can already be seen driving on American highways. In the future, humans will learn to interact with the intelligent creatures in whatever form they come in, from IBM’s Watson supercomputer playing Jeopardy to Manuela’s awkward, but humble CoBot fetching coffee.
“I struggle many times to keep this engineering way and also justify why this is a great theory. In the sense that, I like things that work and sometimes the research is in finding the best way to make them work. But you can not prove a theorem that there exists no other way to do it. I can’t. I just find a way to accomplish the goal. I continue to be curious to be about if there is a better way to do it. It’s not something I can prove analytically. Every year we are doing it better,” Manuela said.
Even though robots are being designed to be more like humans, some human behaviors are more difficult to replicate than others.
“My soccer playing robots know nothing about transporting things. The CoBots know nothing about soccer. They are very specific. My interaction with Cobot is very boring,” Manuela said. “I speak 5 languages fluently, scramble eggs, play squash, and I can do robotic research. Maybe one day a robot will have all of these things, but we are far from that.”
Even though Manuela’s robots may be boring, since they occupy the same spaces that we do, we will have to learn how to interact with them, just as we do with other people and their pets.
Manuela explained, “I think there will be robots, machines of all sorts moving around with people, not just people, dogs, and cats. But people, dogs, cats, and robots.”
Subscribe to Elemental’s YouTube channel
Latest posts by Boonsri Dickinson
- Preparing for the Future of AI, Where Robots Can Learn and Ask Humans For Help! - February 12, 2014