Faculty Recruiting Make a Gift

Taming the Swarm: Control and Design of Multi-Robot Systems

27 Feb
Wednesday, 02/27/2013 11:00am to 12:00pm

Mike Rubenstein
Harvard University
School of Engineering and Applied Sciences

Computer Science Building, Room 151

Faculty Host: Rod Grupen

Abstract: Researchers in the field of multi-robot systems envision groups of robots working together to achieve tasks with more parallelism, adaptability, and fault-tolerance than a traditional single robot. Potential applications exist at many scales: tens of robots monitoring an environment, hundreds of robots automating a large warehouse, millions of nanobots entering and repairing the human body. Along with these new possibilities, multi-robot systems also bring new challenges for control, programming, and design. These systems tend to be decentralized, asynchronous, have unknown or variable numbers of individuals, and varying or even conflicting perceptions of the world within the group. Individuals in these systems are also generally less reliable, accurate, and capable when compared to a single traditional robot.

In this talk, I will describe two control strategies for multi-robot systems that take advantage of the parallelism, adaptability, and fault-tolerance, while addressing the difficulties in controlling such systems. First I will discuss an algorithm which allows a group of robots to autonomously self-assemble any desired connected shape. This shape is formed at a size that is proportional to the number of robots in the group. If the shape is damaged through the removal, addition or re-location of some of the robots, then this algorithm will allow the group to re-form the shape at a size proportional to the new number of robots. Second, I will present a decentralized method by which a group of robots can coordinate and move an object too big for any individual to move. Unlike many other approaches to collective transport, this method does not require robots to know object shape, center-of-mass, attachment points to the object, or the number of other robots. I show that this method is provably guaranteed to move the object in an optimal path to the goal, and that many characteristics such as the scaling of speed with the number of robots can be predicted. This method is validated with extensive experiments, including one demonstration in which 100 robots transported an object, and another in which a "wiggling" object was transported.

Additionally, I will present two examples of multi-robot systems I have helped design. The first example is a modular self-reconfigurable robot called "Superbot". Superbot consists of many individual robots (modules) that are fully autonomous and can attach and detach from each other to form a larger robot. This allows the overall group of Superbots to transform its shape in order to better adapt to changing or unknown environments. The second example multi-robot system I will introduce is Kilobot, a robotic collective of 1024 robots designed to test collective behaviors on a large group of real robots. I will discuss the difficulties in building and controlling such a large group of robots, and how Kilobot addresses these difficulties.

Biography: Michael Rubenstein is currently a postdoctoral fellow in Radhika Nagpal's Self-Organizing Systems Research Group at Harvard University's School of Engineering and Applied Sciences. There he is working on Kilobot, a robot designed for testing swarm algorithms in a group of over a thousand robots. In 2009 he received his Ph.D. from The University of Southern California's School of Computer Science under the supervision of Wei-Min Shen. His thesis, titled: "Self-Assembly and Self-Healing for Robotic Collectives", details a control algorithm for a simple simulated multi-robot system which guarantees that it can self-assemble and self-heal any desired connected shape. Most of his research is centered around the design and control of multi-robot systems. Additional information can be found at his webpage: http://people.seas.harvard.edu/~mrubenst/

A reception will be held at 3:40 in the atrium, outside the presentation room.