Pages

February 03, 2011

Sander Olson interviews roboticist Daniel Lee

University of Pennsylvania roboticist Daniel Lee is a researcher at the Penn's General Robotics Sensing and Perception (GRASP) lab. His students are building robots that compete in robotic soccer leagues, with the ultimate aim being to create a robotic soccer team that can beat the human world cup winners by 2050. Dr. Lee's team has also successfully participated in DARPA's Urban Challenge Program.




Question: Tell us about GRASP lab.

Answer: GRASP is an interdisciplinary center at the University of Pennsylvania that is focused on robotics research. We have students from mechanical engineering, electrical engineering, computer science, and bioenengineering, that are collaborating on various robotics projects, including participating in the Robocup soccer competitions. I am one of the faculty members in GRASP.

Question: There have been Robocup soccer matches since 1997. How much better is the 2011 team than the 1997 team?

Answer: There have been significant improvements in robotics since 1997, so the 1997 robot teams would be no match for our 2011 teams. The initial robots were so bad that it was almost comical - they would lose track of the ball, do nothing, or kick the ball in the wrong goal. By contrast, the 2011 teams are much better. We upgrade the software frequently, and occasionally make hardware changes as well.

Question What specific hardware changes have been made to the robots?

Answer: We started out with four legged robotic dogs, and we have progressed to two-legged humanoid robots. The robots have more sensors and actuators, and the CPUs are faster and have much more memory.

Question: Are there different size classes for robotic soccer?

Answer: Yes, there are two leagues in which we compete. One league uses the commercial small NAO robots without modifications, so the teams have the same hardware and compete on software. The other league is called the humanoid league. The humanoid league, which has 24 teams around the world, is divided into 3 subleagues called the teen size, kid size, and adult size leagues. The humanoid league allows the teams to build their own robots and program them. There haven't been any full adult sized team robotic soccer competitions yet, but I think that we can get there within 5 years.

Question: Why is creating human-sized robots so much more difficult than smaller versions?

Answer: The strength-to-weight ratios of the motors need to increase as size increases. So larger robots require more sophisticated and expensive motors. They also require more energy to run, so the batteries become bigger, and they require more sensors and wiring. We will see increasing numbers of full-sized humanoid robots as the technology improves, but these robots are still quite expensive.

Question What is the ultimate goal of Robocup?

Answer: The short term goal is to use these competitions to encourage robotics research and education for students. The ultimate goal is to have a team of human-sized robots beat the human World Cup winners in soccer. That is an exceedingly difficult goal, but we hope to accomplish this by 2050.

Question: What is the biggest impediment to creating a robotic soccer team?

Answer: The biggest challenge by far is imparting genuine intelligence into our machines. Playing soccer requires quickly sensing and assimilating vast amounts of information, making intelligent decisions, and executing actions. Each one of these tasks in itself is challenging, but imparting general purpose intelligence is clearly the most difficult of those tasks.

Question: What is the status of the Ben Franklin unmanned vehicle racing team?

Answer: We successfully competed in the DARPA urban challenge. The technology that was developed for the DARPA urban challenge is now starting to be commercialized, and Google is now investigating unmanned vehicles. We also have just competed in the MAGIC 2010 competition, and we placed second.

Question:The MAGIC 2010 competition was another DARPA challenge?

Answer: It was an urban intelligence, surveillance and reconnaissance scenario sponsored by the Australian and US Departments of Defense. The final competition was held in Adelaide, Australia, and required teams of robots to search and map a 250,000 square meter area in less than four hours. We succeeded using a team of five sensor robots and two disruptor robots. We won second prize and a $250,000 award, and garnered considerable interest from the military.

Question: Will self-driving cars become commercially available within the next decade?

Answer: That depends upon your definition of "commercially available". The impediments at this point are actually more legal than technical. So we will see more semi-autonomous driving features, in which the car assists you in driving, and this is already happening. The technology needs to be extensively tested and verified before people are going to be willing to have an unmanned vehicle drive their children to school.

Question: Do effective algorithms for machine learning in robotics exist ?

Answer: Yes, and they are being used by companies such as Google and Intel. There are now research robots that can learn about their environment and adapt to it in real time. There is still much room for improvement, and it is an active area of development in laboratories across the country. I am very optimistic as to what is possible - there are many opportunities for continued improvement.

Question: What represents the current state of the art for computer vision?

Answer: We are currently unable to use the most sophisticated computer vision algorithms because they simply require too much processing power for real-time applications. As computational speeds improve, we will see more sophisticated algorithms being employed.

Question: How long before robotic vision equivalent to human vision becomes available?

Answer: Human vision actually requires an extraordinary amount of computing power, as well as understanding fundamental algorithms. It could be at least another decade before machine vision comes anywhere close to matching human vision.

Question: The H-HREX Hexapedal robot has an innovative walking capability. Could this design be copied to larger robots?

Answer: The H-HREX is a robot that moves in a manner similar to a cockroach, using an interesting leg design. I think that locomotion systems based on insects are great for smaller robots, but scaling to larger systems will need further research in innovative leg design, control, and planning.

Question: How long before a thriving commercial robotics industry exists?

Answer: There is a symbiotic relationship between the consumer electronics and robotics industries. The consumer electronics industry has employed the inertial navigation, GPS, and other sensors that were originally used in robotics. This mass adoption has led to massive price reduction in the cost of the sensors. Robotics researchers are in turn taking advantage of these low-cost sensors to design increasingly capable and inexpensive robots. These trends will hopefully continue until the robotics field becomes a thriving, self-sufficient industry. It is exciting to see robotics corporations and labs becoming well-financed, and increasing resources being invested in next-generation robots.







If you liked this article, please give it a quick review on ycombinator or StumbleUpon. Thanks
blog comments powered by Disqus