Here is the Stefan Schaal interview, attached. Dr. Schaal is a roboticist at the University of Southern California who runs the USC Computational Learning and Motor Control Lab. Dr. Schaal has researched many aspects of robotics, including statistical learning, motor control, computational neuroscience, and biomimetic and humanoid robots.
Question: Your lab at the University of Southern California has done research into supervised and reinforcement learning methods for robots. How does this work?
Answer: In reinforcement learning, the robot receives “rewas” for doing tasks properly, or at least for moving in the right direction. Supervised learning provides precise error measurements. So a supervised learning system might inform a robot that their movement was 3 degrees to the left of the object, whereas reinforcement learning would tell a robot that they are getting warmer as they get closer to completing the task.
Question: What does imitation learning entail?
Answer: In imitation learning, a human demonstrates to the robot the task to be performed, and the robot tries to emulate the human action. We can now employ reinforcement learning, supervised learning, and imitation learning to impart almost any technical skill to a robot.
Question: Tell us about the computational brain robot. Was it really controlled by a monkey?
Answer: My collaborators at the Advanced Telecommunications Research Institute (ATR) had a monkey that was directly controlling a robot. The robot moved according to what the monkey was thinking. Initially, the monkey was trained to make the robot walk by seeing the robot’s actions on a computer display. After training, the monkey could make the robot walk without feedback. So it was a proof-of-concept experiment in brain-machine interfaces, and one of the ultimate goals with that line of research is to create human controlled mechatronic limbs.
Question: Your lab has recently made an advance with the LittleDog robot.
Answer: The LittleDog robot is a small robot that can quickly traverse uneven terrain. The real world features uneven terrain, so a robot that can quickly climb over difficult obstacles would be quite useful. The video that we have on our website shows that we have succeeded, and the task now is to impart similar skills to larger, two legged robots.
Question: You are currently working on a humanoid robot. How will this robot compare to Asimo?
Answer: Asimo is a very sophisticated, beautiful robot. But Asimo has a limited ability to interact with the outside world and to react to unforeseen events. If Asimo is pushed, it may fall over and not “gently give in” as a human would do. With the help of Sarcos, we are developing a prototype robot that is going to be higher performance, more adaptable, and more suitable in working in a unpredictable human world than Asimo. This robot will be based on the Sarcos exoskeleton technology, and we hope this humanoid robot technology can be developed over multiple iterations to a model that is even more capable.
Question: Honda’s Asimo robot has garnered considerable attention from the press. Is Japan really the leader in robotics development?
Answer: Japan has the most people working on humanoid robotics. But when it comes to robots that can behave autonomously, the U.S. may have an edge over Japan. So when it comes to really advanced technologies the U.S. and Japan are at par.
Question: Current robots are limited both by insufficient power sources and actuators. To what extent are these problems being remedied?
Answer: There are a number of labs in the U.S. and elsewhere that are working on better actuators and power sources. Fuel cells, for instance, are being investigated as a robotic power source, since they can store more energy than batteries. Hydraulics can be used for muscles on robots, but tends to be heavy and cumbersome. Researchers are currently investigating several different approaches for artificial muscles, including pneumatic artificial muscles, which use compressed air. There is nothing, however, to compare to human muscles, which are marvels of speed and efficiency.
Question: Computers are also limited by computing power. Is anyone researching using GPUs to address this problem?
Answer: GPUs are becoming increasingly popular in robotics. For any problem that requires massive parallelism, GPUs are an excellent choice. The biggest problem now is learning how to program the robots. Once we can create the software, we can create specialized chips that efficiently perform the task in hardware.
Question: So software is more of a problem than hardware?
Answer: The software is not good enough yet — essentially, humanoid robotics needs to address a very large scale artificial intelligence problem, which, so far, is unsolved. Once it is, we can work with the hardware developers to realize the software with specialized hardware.
Question: Is funding for robotics projects increasing? It seems that an increasing number of projects are getting funded.
Answer: Yes, robotics is experiencing a new wave of funding from both Government and commercial sources. It is only a matter of time before robots become ubiquitous, and countries don’t want to miss out on the coming revolution. So we are seeing robotics programs proliferating in the U.S., Europe, and Asia.
Question: What advances will be needed before robots can operate in an unstructured environment?
Answer: There are two deficiencies that are impeding robotics development. First, we need better actuators, since actuators available now simply aren’t up to the task. We also need better tactile sensors, in order to let robots interact with their environment. Ideally, a robot would have a sensor capability as robust and comprehensive as a human. The problem with sensors is that as the number of sensors increases, so does the wiring and demands on signal processing.
Question: How can this wiring problem be addressed?
Answer: There are a number of ways to address this problem. One approach is to use wireless sensors. Another technique is to use sensor networks, in which sensors only communicate with their nearest neighbors. This would involve putting as much computational power in the sensors as possible. Yet another possibility is to multiplex sensors, which significantly reduces the wiring required. So this issue is tractable, and will soon be surmounted.
Question: Microsoft is giving away its robotics developer studio software. Willow Garage is giving away its PR2 robots. Is robotics development now largely open source?
Answer: I haven’t seen a large scale adoption of Microsoft’s developer studio software yet. Willow Garage is creating an open-source software model, and is also giving away its PR2 robots. Instead of creating products that are incrementally superior, Willow Garage is developing the underlying technologies that a robotics industry will require. If Willow Garage succeeds, it will exponentially increase the pace of robotics development.
Question: What specific robotic advancements do you anticipate for the next two decades?
Answer: Within the next 20 years, we will get to the point that an individual can purchase a general purpose robot. The individual would then tell the robot what it wanted to do, and the robot would learn to perform the required task on its own. By 2030, robots should be able to function safely, effectively, and somewhat autonomously in an open, real-world, unstructured environment to perform simple tasks. The robotics industry is starting its exponential growth phase, and this phase should last for at least several decades. At some point in this century, robots will be ubiquitous in society, and will play key roles in assisting humans in their daily activities, ranging from entertainment, health care and education to labor and finally potentially dangerous and hazardous tasks.
Brian Wang is a Futurist Thought Leader and a popular Science blogger with 1 million readers per month. His blog Nextbigfuture.com is ranked #1 Science News Blog. It covers many disruptive technology and trends including Space, Robotics, Artificial Intelligence, Medicine, Anti-aging Biotechnology, and Nanotechnology.
Known for identifying cutting edge technologies, he is currently a Co-Founder of a startup and fundraiser for high potential early-stage companies. He is the Head of Research for Allocations for deep technology investments and an Angel Investor at Space Angels.
A frequent speaker at corporations, he has been a TEDx speaker, a Singularity University speaker and guest at numerous interviews for radio and podcasts. He is open to public speaking and advising engagements.