Minimal Information in Sensing and Control argues that with limited sensing and control, very interesting control tasks can be achieved without state estimation. We successfully proved that one such task, rendezvous, can be accomplished by using couple detection beams, three-level quantized control for a group of Dubins car (a type of differential drive) agents.

Related Publication(s):

Rendezvous without Coordinates. J. Yu, S. M. LaValle and D. Liberzon. Submitted to IEEE Conference Decision and Control, 2008.

Abstract: We study minimalism in sensing and control by considering a multi-agent system in which each agent moves like a Dubins car and has a limited sensor that reports only the presence of another agent within some sector of its windshield. Using a very simple quantized control law with three values, each agent tracks another agent assigned to it by maintaining that agent within this windshield sector. We use Lyapunov analysis to show that by acting autonomously in this way, the agents will achieve rendezvous if the initial assignment graph is connected. A distinguishing feature of our approach is that it does not involve any estimation procedure aimed at reconstructing coordinate information. Our scenario thus appears to be the first example in which an interesting task is performed with extremely coarse sensing and control, and without state estimation. The system was implemented in computer simulation, accessible through the Web, of which the results are presented in the paper. [pdf] [ps.gz] [Web Demo]

Shadow Information Spaces research reasons around the evolvement of information that's fully or partially hidden to the robots' sensors. We argue that given the transitioning of shadow components, the problem of estimating those hidden information can be transformed into a maximum flow problem on a bipartitie graph structure. Application includes tracking, counting, etc.

Related Publication(s):

Tracking hidden agents through shadow information spaces. J. Yu and S. M. LaValle. Accepted by Proceedings IEEE International Conference on Robotics and Automation, 2008. Also in Technical Report UIUCDCS-R-2007-2902, Department of Computer Science, University of Illinois, September 2007.

Abstract: This paper addresses problems of inferring the locations of moving agents from combinatorial data extracted by robots that carry sensors. The agents move unpredictably and may be fully distinguishable, partially distinguishable, or completely indistinguishable. The key is to introduce information spaces that extract and maintain combinatorial sensing information. This leads to monitoring the changes in connected components of the shadow region, which is the set of points not visible to any sensors at a given time. When used in combination with a path generator for the robots, the approach solves problems such as counting the number of agents, determining movements of teams of agents, and solving pursuit-evasion problems. An implementation with examples is presented. [pdf] [ps.gz]

MS/PhD in Computer Science, University of Illinois at Urbana-Champaign, 2006 - ?

M.S. in Mathematics, University of Illinois at Chicago, 2001

M.S. in Chemistry, PhD candidate, University of Chicago, 2000

B.S. in Materials Science and Engineering, USTC, Hefei, China, 1998