2011: Baby Steps to "The Jetsons"
When people find out that I work in the personal robotics industry, they invariably ask me how quickly their Rosie the Robot will arrive. Everyone wants his/her own personal assistant. We’ve already seen robots tackling laundry, sorting socks, and even fetching a beer; so it’s fair to say that a good deal of progress has been made toward a future where personal robots are routinely assisting humans.
The idea of personal robots routinely in the home or office, however, is still far off. So just how close are we? “The Jetsons” takes place in the year 2062, and we just might beat that date (minus Rosie’s famous attitude)–but there are still some significant challenges around a robot’s functional capabilities; namely, perception, planning, and manipulation.
Perception simply means a robot seeing our world and understanding what’s in it. We have a way to go here. Our PR2 robot, for example, views the world through the lens of space and points. It is programmed to navigate through space and sense points where light reflects, fundamentally creating a point cloud model of the world. Today, a human being and a mannequin share the same characteristics with respect to size and shape–at least as far as a robot is concerned. Significant improvement is needed here, but unsurprisingly, it’s where a lot of research is currently focused.
Planning is already solved, as long as the robot stays in two dimensions. But if you are trying to move a robot arm, which can typically move in six different ways (e.g. elbow bend, shoulder bend, forearm twist, etc.), each of those joints may as well represent a dimension. Think about the simple act of taking a sip from your coffee cup. The human brain makes instantaneous calculations on weight, temperature, weight distribution, and more. All of this before the first taste. Robotics has made great progress in this type of planning, but when you see a robot performing tasks such as this one, it is slow and clunky. We’ve seen algorithms come out in the past year that have improved this process, but we’ll need much more improvement in this capability.
The last challenge, manipulation, is more than just recognition and reaction. If perception has correctly identified an object to be manipulated the robot may be able to look it up in a database to find out the best place to grasp the object securely, but that doesn’t provide common sense. For example, when clearing a table, there’s more to picking up a partially-full soup bowl than knowing how to grasp the bowl so you don’t drop it: the robot must recognize that there is liquid in the bowl and keep it upright, and make sure the spoon doesn’t fall out during transit.
These challenges are currently being tackled in a variety of ways. One is to better prepare the robot to deal with corner cases–instances in which it maneuvers itself into a corner, either literally or figuratively, and must maneuver its way out. Another is to build a communication infrastructure in which humans are easily inserted into a robot’s decision-making capabilities. Robots don’t have common sense, so they need to be taught how to deal with exceptions. That education can be as simple as text messages, for example, with the implication that the message will grow the robot’s local knowledge.
In 2011, expect to see more Human-in-the-Loop (HITL) modeling and simulation. Humans are amazingly efficient at perception. Activities we take for granted are still far too slow and inefficient on a robot. By using the knowledge of humans around the world (crowdsourcing), we bring the potential to provide a two-way line of communication between the robot and a person who is interacting with it, without needing to solve artificial intelligence.
I have two other expectations for 2011. One is the ongoing growth of ROS, an open source Robot Operating System. It’s already on an incredible exponential growth tear, but we are now starting to see the network effect kick in. Even since our State of ROS post back in November, adoption has grown by roughly another 20 percent. It’s reliable, it’s stable, it’s free and it’s the most efficient means to accelerate innovation.
Expect to see more sophisticated low-end robots this year. Until recently, the world of low-end and hobbyist robotics has been sensor-deprived. This is why so many people are excited about Kinect. Now you can combine an iRobot base, a laptop and a $150 Kinect and suddenly you have a robot that is under $1,000 and has a 3D view of the world. These robots now have the means to navigate the world in a way that previously simply wasn’t possible.
We’ve built the Willow Garage business on the proposition that the personal robotics industry will only truly begin to mature when common hardware, software and best practices accelerate innovation and collaboration. The personal robotics industry has never had a common operating system such as ROS before; or a state-of-the-art R&D robot such as the PR2. As the price of personal robots continues to drop, and the software and hardware capabilities continue to improve, 2011 feels like a bellwether year for personal robotics. It may not be the year that Rosie arrives on your doorstep, but she’s well on the way.
Prior to joining Willow Garage, Steve was the senior manager of the User-Focused Systems Research Group at the IBM Almaden Research Center, one of the top human-computer interaction research groups in the world. Earlier, Steve managed the Advanced Systems Development Laboratory at the Xerox Palo Alto Research Center (PARC).
