Zack Butler's research

My general area of interest is cooperative robot systems, particularly self-reconfigurable robots, and modeling and control of cooperative natural systems. Below are short descriptions of projects I am actively working on, though I am always interested in listening to project ideas in robotics and cooperative systems.

Self-reconfiguring robotics

Self-reconfiguring robots are systems made up of a large number of simple modules which can form themselves in to arbitrary shapes to accomplish different tasks. The primary basic functionalities are locomotion (moving the system across the landscape), in which a group of modules may change from a snake-like shape to fit through small tunnels to a tall scaffold-like shape to surmount tall obstacles, and reconfiguration, in which the modules are specified to achieve a given exact shape, such as for 3-D visualizations. My current research includes two directions primarily focussed on the former task:

Reconfigurable Teams
In this work, I am interested in using the power of self-reconfigurable robots to divide and spawn new independent robots during task execution. Essentially, this can be thought of as an extension to multi-robot task allocation, in which the number of robots and their capabilities is no longer fixed. The question of how to optimize robot actions, especially in a distributed context (both modules within connected groups and groups within an overall system) is a challenging one with many possible answers. Our first foray into this arena was with a purely reactive algorithm for reaching given goal locations in an unknown environment (see our DARS paper). Currently we are looking into optimal tours in known environments (sort of akin to k-TSP), developing reactive sensor placement algorithms, and applying various multi-robot auction algorithms to such systems.
Efficient locomotion algorithms(with R. Fitch, Australian Center for Field Robotics)
The reconfigurable teams work assumes that each group of modules can move about its environment as needed, and various groups have developed algorithms for different systems to allow these capabilities. In our latest contribution to that part of the field, we are developing simple decentralized algorithms that can perform locomotion of a group over arbitrary terrain. The focus of this work is on minimizing the computation and memory requirements for large (e.g. 1 million module) systems. See our ICRA paper. We represent the problem as a Markov Decision Process in which the local values of the MDP are kept locally within each module, i.e. a constant amount of memory per module. Updates are performed via message passing among the modules, and each module can query local MDP values to determine its best move. Accompanying this is a novel motion coordination algorithm that ensures modules can move safely while keeping coordination messages to a minimum. We are currently working on extending this technique to other types of actuation in the hope of implementing it on hardware, and eventually integrating it with the reconfigurable teams work to show a complete (simulated) system performing a set of tasks in parallel.
Hierarchical simulationsMy latest area of interest ties in both of the preceding bullets - producing algorithms and simulations that handle the inter-module decisions and inter-blob decisions simultaneously. I am especially interested in creating simulations that run efficiently over a cluster, so if you are similarly interested, please get in touch!

Computational Tools for Managing Herds (with D. Rus, MIT and D. Anderson, USDA)

In this project, we are trying to bring modern computational tools to the field of livestock management. In particular, we want to use computing devices along with GPS and stimulus to provide "virtual fences" for animals in the field. These fences can be used to maintain the herd in an arbitrary portion of the overall rangeland, eliminating the need for large amounts of expensive and time-consuming fence maintenance. Even more powerfully, however, we believe that these fences can be moved over time to produce mustering or other herd motion. This opens the door to using a wide variety of algorithms (many of them inspired by similar problems in robotics) to control the motion of the animals over the landscape. One algorithm I developed is a novel path planner to direct a group from one point to another with minimal intervention through the use of a virtual corridor (see paper). We have also looked into using such corridors as a way to control a team of simple robots, essentially herding them instead of requiring them to travel in a particular formation (which they may not be capable of depending on their sensing abilities). I am also interested in the problem of coverage, moving the animals in such a way as to best utilize the available land, neither overgrazing nor undergrazing to any great degree. However, it is also important to do any such control in a way that works with the animals' natural tendencies rather than against them. Greg von Pless developed a novel adaptive expert system that decides how to move the cows around their environment, and we have published this technique in an ICRA paper. This strategy is rather proactive in that it attempts to keep the animals in a given area and move the area around - we are also looking into more reactive strategies that will presumably allow the animals to have more natural behaviors but will hopefully also minimize over/under-grazing.

Using mobile robots to teach advanced (non-robotics) computing topics (with R. Raj and M. Kwon, RIT)

We are interested in bringing the excitement and immediacy of robotics to topics that traditionally have used only software approaches in their instruction, such as peer-to-peer databases and mobile ad-hoc networks. To this end, we are hoping to obtain a number of simple robots with which we will provide simple motion functionality. Students in a course can then add their algorithms on top of the moving platform without worrying about the robotics aspects, and see how real-world effects intrude on the execution of various algorithmic approaches. More details on the current progress, robots, and so forth can be seen at the lab web page.

zjb-AT-cs-DOT-rit-DOT-edu