January 21, 2010

James Gunderson, of Gamma Two Robotics interviewed by Sander Olson

James P. Gunderson, Ph.D., interviewed by Sander Olson

Dr. James Gunderson is the Systems Architect for Gamma Two Robotics. He has been actively researching robotics for the past two decades. Gamma Two Robotics has created two robots with "cybernetic brains" - computers that allow the robot to react and alter their behavior based on various environmental stimuli. These robots are equipped with sonar and will soon also have vision and hands, and will eventually be able to rewrite its algorithms and continuously learn. Gunderson believes that these "cybernetic brains" could be the first step toward artificial general intelligence

Gamma Two Robotics at CES 2010 Video

Question 1: Your company has developed two new robots, the Basic Service Level (BSL) model and the Wheelchair Level Mobility Assistant (WLMA). What is special about these robots?

Answer: Unlike most robots, our robots actually know what they are doing. Most industrial robots only follow preprogrammed courses. Even the romba robot cannot discern between a person and a chair. By contrast, our robots are constantly sensing the environment and building symbolic models of the environment. So they can alter their behavior based on their environment, much like a living system does.

Question 2: How do the robots sense their environment?

Answer: They make extensive use of sonar and thermal sensors, so no vision is currently involved. So these robots can not only differentiate between a person and a chair, but they can determine whether their chair has 4 legs or if it has wheels on the bottom.

Question 3: What is the "cybernetic brain" that powers these robots?

Answer: The cybernetic brain was designed by my wife, Dr. Louise F. Gunderson. She has a Ph.D. in Systems Engineering, and she and I examined how living organisms model their world. The cybernetic brain is directly based on this modeling.

Question 4: This brain sounds like an AI advance. Is there any chance that this "cybernetic brain" could be the first step towards artificial general intelligence?

Answer: Absolutely. The mechanisms in the brain are very generalized, and we designed our system to emulate the pre-frontal cortex of the human brain, the part that "thinks about thinking". So the cybernetic brains are generalized modules that are capable not only of learning, but of continuously altering their behaviors.

Question 5: To what extent can these robots learn?

Answer: Our robots incorporate three distinct types of learning. They can learn positionally - we are doing a demo here at CES 2009 in which we take a robot and teach it several new locations. As soon as it learns these it knows how to move between them. It also has an episodic memory that records every action that it takes and every decision that it makes and can recall that memory. The third component is its ability to rewrite its algorithms and restructure how it perceives the world. This third component does not yet exist, but this should be developed within the next decade.

Question 6: Do you have plans to add vision to these robots?

Answer: Yes, vision will be added during the next 18 months. Given the sensitivity of our sonar, there are a large number of tasks for which vision is unnecessary. But in the long run vision is essential, so we plan on incorporating vision into all of our models. Future models will include both vision and sonar.

Question 7: Several robotics companies have unveiled sophisticated robotic hands. Will Gamma Two Robotics license this technology and incorporate it into the BSL and WLMA models?

Answer: We are amenable to adding hands and arms to our models, and it should not be too difficult to do this. The difficult part is imparting fine-grained control to the limb, and that involves making new classes for the brain to inform it that it has the limbs to use. It also requires vision. But there are some wonderful robotic arms and hands available now, and we are eager to license them once we have upgraded our cybernetic brains to recognize and use them.

Question 8: Your robots currently lack the ability to open doors or climb stairs. How difficult will it be to add these capabilities to future robots?

Answer: There is actually a DARPA grant for robots to open doors. It is actually surprisingly difficult for a robot to open a door. But there are a number of University programs that are tackling this issue, and I'm confident that robots a decade hence will routinely be able to open doors and climb stairs.

Question 9: These robots can respond to verbal commands. How many commands can they respond to? Can they speak?

Answer: The entire interaction with our robots is via voice-input and voice-output. These robots have a fairly simple command language - they aren't doing natural language processing. They understand words like "go to" or "deliver" or "move" or "turn". They also have a semantic memory - the part of a living system that remembers things like "cars are things that drive on roads" and "chairs are things that you sit in". So interaction with these robots is both natural and straightforward.

Question 10: Both of your robots have wheels in place of limbs. Have you ever considered building a robot with legs?

Answer: One of the main problems with legs is power - legs require substantially more power to operate than wheels. Legs need not only to move forward but also to keep the robot off the ground. Our wheeled robots can run for four hours, but if we tried to put legs on the robots the run time would drop dramatically. The advantage to legs is that human spaces are designed for legged individuals. But since we are looking at a primary market for individuals with limited mobility, the limited mobility of our robots isn't a major problem.

Question 11: The U.S. military is investing heavily in robots. Do you have plans to develop any military robots?

Answer: We don't have plans to develop any military robots. Most of the military's robots are under operator control at all times, and robots just aren't sophisticated enough at this point to autonomously control weapon systems. Our primary focus is making robots that relieve workloads. We are looking at a program with the veterans administration for mobility assistance for veterans, but we have no plans for a combat robot.

Question 12: You have been doing robotics work for the past ten years. Is the pace of robotics development greater now than it was in 2000?

Answer: I don't think it is greater. The robotics programs in Academia have been fairly constant for the past 30 years. We have seen incredible progress on the hardware side of things, such as Honda's Asimo, and Boston Dynamics Big Dog. But what he haven't seen before now is the ability to move beyond preprogrammed or purely reactive motion. That is what makes our robots special.

Question 13: What field will be the biggest driver of robotics during the next decade?

Answer: I think the biggest driver for robotics during the next decade will be in the home health assistance field. With the burgeoning number of elderly people and the chronic and worsening shortage of workers to care for the elderly, this field could be transformed by robots. In the immediate future we are seeing demand for our robots from catering companies. Robot catered parties and events is a major draw for people.

Question 14: If Gamma Two Robotics succeeds as you hope, what types of robots will your company be offering in 2020?

Answer: By 2020 we plan on offering a full range of semi-autonomous robots for a wide variety of circumstances - robots to help the handicapped, robots for catering, and robots for home and office markets. To give one specific example, robots could be accompanying people to the grocery store and helping them shop. These robots will be able to function in the real world, will come standard with arms and vision, and will be able to follow general commands. They could become as indispensable in 2020 as cell phones are today.

Video interview with Dr. Louise Gunderson

blog comments powered by Disqus