Here is the David Barrett interview, which was conducted by Sander Olson. Dr. Barrett is a Professor of mechanical engineering and design at Olin College who teaches robotics. He heads Olin’s Senior Capstone Program in Engineering (SCOPE) program, which is designed to introduce students to robotics research. Dr. Barrett received his PhD from MIT, and has worked at the irobot corporation, the Draper Laboratory, and MIT’s Artificial Intelligence Laboratory. Dr. Barrett holds nine patents on robotics and is considered an expert in the field.
Question: Olin College’s Senior Capstone Program in Engineering (SCOPE) seems specifically designed to promote robotics development.
Answer: SCOPE is a very hands-on engineering curriculum for our students. Freshman students are put on teams that are given small tasks to complete. By the time they are seniors, our students are put on major projects that are sponsored by corporations to the tune of $50-75,000. SCOPE is heavily oriented towards robotics applications, and we currently have 5 robotics projects underway.
Question: Your students have just created a robotic fish.
Answer: The fish is designed to swim in the water in a manner similar to actual fish. It is actually designed for navigation in very crowded aquatic environments, such as streams and rivers. One of the primary uses of this fish is to monitor marine environments, so I could envision thousands of these robotic fish being used to continuously monitor and explore the oceans.
Question: Your students are also in the process of creating an unmanned driving vehicle. What capabilities will this vehicle have?
Answer: It is designed to drive autonomously at relatively high speed through unstructured off-road environments. Driving off-road is actually more difficult than driving in traffic. Off-road presents many of the same challenges as regular traffic but also presents other 3-d obstacles as well. This vehicle should become fully operational by May 2010 and will be capable of point-to-point navigation using GPS. Or you could teach it a route by having it drive that route once and it will automatically learn the route.
Question: Could this car be used as a personal chauffer?
Answer: Potentially, but that is not its primary market. Its more immediate market is in farms and construction. Many farms and construction sites are manpower limited, and would directly benefit from having automated vehicles that could move cargo.
Question: You have worked at MIT’s AI laboratory. Are you anticipating any AI advances that could benefit the robotics industry?
Answer: We are building machine learning into both the unmanned vehicle and the robot fish. By studying the robots performance in complex terrain their control systems are actually continuing to improve. I predict that the machine learning part of robotics will have a huge payout in the next decade.
Question: What sort of machine learning improvements are being made?
Answer: Much of the improvement pertains to pattern recognition. For instance, our robotic fish is being guided around the waters at test stations. The machine-learning engine is doing pattern recognition. So it can learn to navigate on its own based on observing a human being navigate it. It learns on its own, and continually self-optimizes.
Question: To what extent is information overload a problem for robots?
Answer: Information overload hasn’t been a problem for us yet. We primarily use 3-million gate Field programmable Gate Arrays (FPGAs) that have phenomenal data-capturing capabilities. Within the next several years, robots equipped with vision, sonar, radar, and other sensors should become commonplace. The more extensive the sensor suite, the more useful the robot becomes.
Question: But increasing the sensor inputs will inevitably increase processing requirements.
Answer: Yes, but biological systems make extensive use of parallel processing, and we are increasingly doing the same with robots. Robotics is a field well-positioned to take advantage of many-core processing. We are also starting to use GPUs for processing, and I foresee combinations of FPGAs and GPUs meeting the increased processor requirements of future robots.
Question: What industry will be the biggest consumer of robotics during the next decade?
Answer: I see the medical industry being a major driver of robotics during the next decade. The medical field is not nearly as automated as it could be, and I foresee large numbers of laboratory technicians being replaced by robotics systems. Genomics is already highly roboticized, and provides an excellent example of the efficiencies that can be gained from automating routine tasks.
Question: What about robot orderlies?
Answer: There is a substantial amount of R&D going into that area. But the ability of a robot to interact with a human is still limited, and robot limbs are not sufficiently safe to allow them in close proximity to humans. So this market area will probably be huge eventually, but not in the next decade.
Question: When do you think the first robotic equivalent of the Model-t will be made?
Answer: It all depends on the industry. In UAVs, we are already well past the Model-T stage. The irobot vacuum cleaner is already at or beyond the model-T phase. We are probably only a few years away from sophisticated robot toys that could be programmed for various tasks. An autonomous robot butler/servant, however, is probably several decades away.
Question: How sophisticated will robotic vision need to be before robots can safely function in real-world environments?
Answer: Robotic vision needs to be faster and more sophisticated before it can handle real-world environments in real-time. We are probably five years away from a vision breakthrough. The combination of coupling GPUs with low-cost cameras could dramatically improve robotic vision. Unlike radar and sonar, which are inherently complex and expensive, robotic vision is going to become ubiquitous.
Question: How long will it take for speech recognition robots to become commonplace?
Answer: Speech recognition is already successfully used in laboratory conditions. We are developing robots that can respond to commands in plain English, in particular a robotic chauffer. Within a decade, such robots should be commonplace.
Question: If you had a billion dollars to invest in robotics research, how would you allocate your funds?
Answer: I would allocate the money to four particular areas that would have the greatest impact on robotics development. These four areas would include machine learning, vision systems, better actuators, and better sensors. I don’t foresee any showstoppers in any of those areas, given sufficient funding. If enough smart people work on these problems, rapid progress will be made in all of these fields.
Question: Will the next major big robotics advance come from two researchers in a garage, or from a large R&D lab?
Answer: I expect the next major robotics advance to come from a University. Colleges and Universities have a nearly optimal combination of researchers, talented people, and infrastructure. I envision a situation where several researchers start out developing a robotics technology at a University lab, and then move to a garage and start their own robotics company.
Question: How rapidly are the number of robotics companies increasing?
Answer: There are already close to 80 companies in Massachusetts alone that claim to be robotics companies, and that number is growing each year. This trend mirrors that of the personal computer industry growth in the late 1970s. So we are entering the knee of the exponential growth curve for this technology.
Question: What specific developments would you like to see in the field of robotics during the next decade?
Answer: I would like to see FPGA and GPU based robotic brains. We are hitting a wall with conventional CPUs, because classic CPUs aren’t well suited to robotics applications. But FPGAs and GPUs are inherently parallel systems, and so are nearly ideal for robotics applications. Better actuators, equivalent in performance to human muscle, would also be enormously helpful. A lightweight, high-powered, easy to control linear actuator would open up a huge range of applications, and provide a huge boost to the robotics industry.
If you liked this article, please give it a quick review on Reddit, or StumbleUpon. Thanks