I am fascinated by the various ways that the simple local interactions of many achieve complex behavior in a global scale.
Currently I pursue my Ph.D. at Koen Langendoen's Embedded Software group, Software Technology, Delft University of Technology, Netherlands.
We consider the problem of navigating a mobile node to the extremum of a scalar field that is monitored by a wireless sensor network. We are motivated by the scenario of a festival, where the task is to monitor and manage crowds of people. In this scenario, every participant is equipped with a sensor node in the form of a wrist band. The devices form a wireless network, which allows to monitor the density of the crowd. In cases of panic, we want to navigate each attendee away from areas of extreme crowd densities where dangerous phenomena may take place. Due to the high mobility, and low cost and power constraints of the wristbands, no location information can be used. We address this requirement by using gradient navigation methods.
This was the result of Ioannis Protonotario's thesis work.
The decentralized detection of event regions is a fundamental building block for monitoring and reasoning about spatial phenomena. Our study proposes a theoretical framework with which we can analyze event detection algorithms suitable for large-scale mobile networks. Our analysis builds on the following insight: the inherent trends of spatial events are well captured by the spectral domain of the network graph.
Using this framework, we propose (four) novel local algorithms that are location-free; that work with mobile nodes and dynamic events; that operate on 3D topologies; and that are simple to implement.
Above, we show a short demonstration of our most effective detector. Green nodes identify the presence of an oil spill (dark gray shape) out of a noisy environment (light gray background). The datasets are forecasts of the Deepwater Horizon oil spill. Note how the algorithm reconverges instantaneously when the oil and non-oil regions are inverted (10 sec into the video). On the right, we plot the instantaneous values of two detection metrics: the true positive rate (above) and the false positive rate (below). Ideally, true positive rate = 1 and false pasitive rate = 0.
Information potentials can be thought of as spatial aggregates of a quantity distributed over a network. The computing local algorithm distributedly generates a spectrum of possible information potentials between the extreme points of a local view and distributed averaging.
We here observe the remarkable property of potentials to be unimodal (i.e., feature a single, global minimum) under certain parameter ranges. Unimodality is a very valuable property for chemotactic search, which can be used in diverse application tasks such as directed search of information and rendezvous of mobile agents.
In this example, the quantity in consideration is node density. The network consists of 1000 nodes, walking according to the random direction mobility model in a 1x1 km playground. The simulation evolves ten times faster than reality, i.e., one video second corresponds to 10 simulation seconds.
I have also made a strip-down version of the simulator available for experimentation. Extract, and execute as follows:
java -jar potentials.jar network.png
While the simulation is running, you have the option of changing two critical simulation parameters, the inhibiting factor and the communication radius.
Crowds often exchibit complex self-organizing behavior. Below you can observe the result of 1000 people evacuating a hall. On the right I compute the evacuation flow, which is the crowd density multiplied by crowd speed per unit space.
The mobility model was based on the work by Burstedde, et.al. The crowd exists as autonomous entities or agents of a cellular automaton. Its primary objective is self-preservation, thus reaching the door on the right. Like ants, agents, i.e., people, tend to follow each other by emitting and following virtual pherormones. We also take the notion of personal space into account, such that people tend to avoid standing very close to each other.
fair rate allocation
Below you can see a decentralized adaptive algorithm that attains fair rate allocations by operating based on local aggregates. In short, the objective of each node is to claim its fair share (green disk) of the local bandwidth (outer disk). The channel load is shown by the size and colour of the outer disk: white is good, red is bad. While it is not optimal, our algorithm is inherently robust to message loss and incomplete information; thus it is ideal for mobile and crowded networks.
Nodes move according to the random waypoint model. Parameters are set to mimic human runners. The mobility traces were generated with the BonnMotion tool.
courses & offered thesis
Contact me if you are interested in doing a thesis on the following topics: decentralized algorithms with an emphasis on local algorithms for mobile and robotic networks, dynamical systems, local signal processing. The underlying theme is "understanding how local decisions shape/capture global trends".
Embedded Real-Time Systems Course (In4073).
Wireless Sensor Network Seminar (In4316).
A. Loukas and M.A. Zuniga and I. Protonotarios and J. Gao (2014). How to identify global trends from local decisions? Spatial Event Detection on Mobile Networks. In 33rd Int. Conf. on Computer Communications (INFOCOM), Toronto, Canada. IEEE. [ BibTeX ]
M.Cattani and M.A. Zuniga and A. Loukas and K.G. Langendoen (2014). Lightweight Neighborhood Cardinality Estimation in Dynamic Wireless Networks.. In 13th Int. Conf. on Information Processing in Sensor Networks (IPSN), Berlin, PA. ACM/IEEE. [ BibTeX ] (best IP-track paper runner-up )
A. Loukas and M. Woehrle and M.A. Zuniga and K.G. Langendoen (2013). Fairness for All; Rate Allocation for Mobile Wireless Networks. In 10th Int. Conf. on Mobile Ad-hoc and Sensor Systems (MASS), pp. 154--162, Hangzhou, China. IEEE. [ BibTeX ]
A. Loukas and M.A. Zuniga and M. Woehrle and M.Cattani and K.G. Langendoen (2013). Think Globally, Act Locally: On the Reshaping of Information Landscapes. In 12th Int. Conf. on Information Processing in Sensor Networks (IPSN), Philadelphia, PA. ACM/IEEE. [ BibTeX ]
A. Loukas and M. Woehrle and P. Glatz and K.G. Langendoen (2012). On Distributed Computation of Information Potentials. In 8th Int. Workshop on Foundations of Mobile Computing (FOMC), Madeira, Portugal. ACM. [ BibTeX ]
V.G. Iyer and A. Loukas and S.O. Dulman (2012). Nest : A Practical Algorithm for Neighborhood Discovery in Dynamic Wireless Networks using Adaptive Beaconing. Technical Report ES-2012-01, Delft University of Technology, 2012. [ BibTeX ]
A. Loukas and M. Woehrle and K.G. Langendoen (2011). On Mining Sensor Network Software Repositories. In 2nd Int. Workshop on Software Engineering for Sensor Network Applications (SESENA), pp. 25--30, Waikiki, Honolulu, Hawaii. ACM. [ BibTeX ]
O. Akribopoulos, D. Bousis and D. Efstathiou and H. Koutsouridis and M. Logaras and A. Loukas and A. Nafas and G. Oikonomou and I. Thireou and N. Vasilakis and P. Kokkinos and G. Mylonas and I. Chatzigiannakis (2008). A Software Platform for Developing Multi-player Pervasive Games Using Small Programmable Object Technologies. In 5th Int. Conference on Mobile Ad-hoc and Sensor Systems (MASS), pp. 544--546, Atlanta, Georgia. IEEE. [ BibTeX ] (best demo paper award)
I have been very fortunate to work with bunch of wonderful people. Here are some of them:
- Marco Zuniga, Assistant Professor
- K.G. Langendoen, Professor
- Matthias Woehrle, Bosch Research
- Jie Gao, Associate Professor
- Marco Cattani, the author of Fringe Technology
- V.G. Iyer, post-doctoral researcher, SICS
- Andrei Prutaneau, post-doctoral researcher, TUDelft
- Ioannis Chatzigiannakis, Secretary of the European Association for Theoretical Computer Science
- Stefan Dulman, CEO of Hive Systems
- Phillip Glatz
- Andrea Simonneto, post-doctoral researcher, TUDelft.
The D2S2 project aims at developing a framework for programming and operating distributed sensor systems that can be depended on in practical application scenarios. To make an experimental approach feasible, the project focuses on localization and tracking systems in two scenarios that are very relevant to the Dutch society: traffic monitoring and control (static setup) and rescue operations by firefighters and policemen (dynamic setup). A key, innovative feature of the project is the development and use of an advanced miniaturized radar sensor that can operate under a wide range of difficult environmental conditions (smoke, fog, etc.) that cannot be handled by typical localization systems in operation today.
The aim of this project is to provide a multi-level infrastracture of interconnected testbeds of largescale wireless sensor networks for research purposes, pursuing an interdisciplinary approach that integrates the aspects of hardware, software, algorithms, and data. This will demonstrate how heterogeneous small-scale devices and testbeds can be brought together to form well-organized, large-scale structures, rather than just some large network; it will allow research not only at a much larger scale, but also in different quality, due to heterogeneous structure and the ability to deal with dynamic scenarios, both in membership and location.
Office HB 09.030, EWI, TU Delft
CSS Courtesy of Open Web Design