Fluxotaxis for Chemical Plume Tracing

Chemical plume tracing has a wealth of applications, both military and civilian. For example, facilities producing and/or using toxic chemicals are concerned with hazardous leakage, environmental pollutants can cause local concerns, and most recently there have been increasing concerns about terrorists releasing chemical or biological weapons. Our current approach addresses the issue of chemical plumes, but we will soon be addressing biological hazards as well.

The chemical plume tracing (CPT) task consists of three phases: find the plume, trace it to its source and, finally, identify the emitter that is the source generating the plume. Our project is designed to address all three phases. The approach that we adopt is based on the physicomimetics (also called artificial physics, or AP) framework for distributed control of mobile robots. This framework is an efficient, robust, fault-tolerant, scalable, and effective method for achieving desirable behavior with swarms (large numbers) of vehicles. It is based on well-understood physical principles and has been successfully implemented on collections of robots, although it is easily extendible to air or water vehicles as well. Using AP, hundreds of simulated (and several actual) vehicles self-assemble into hexagonal and square lattices (used as a sensor network, or a distributed antenna).

Using these lattices as computational grids, we address the CPT task with a computational fluid dynamics (CFD) approach. Each robot acts as a CFD grid point, thereby enabling the robotic lattice to behave as a distributed computer for fluid computations. Although robots only gather information about the fluid locally (i.e., their own readings and their immediate neighbors' readings), the desired direction of movement for the overall lattice emerges in the aggregate, without any global control. On a set of realistic flow fields, our approach, called fluxotaxis, outperforms the leading competitors for this task. Furthermore, to the best of our knowledge, ours is the first approach to this task that is based on principles of fluid dynamics and has mathematical assurances regarding its success as a local navigation strategy.

Team Members

Results

This 2D simulation of a meandering, turbulent plume shows the temporal and spatial development of a chemical cloud (bright green areas denote high density, and low chemical concentration is shown with dark blue), which originates from the emitter (marked by a small green triangle) located in the bottom half of the simulated environment. The plume is driven away from the emitter by a changing wind (shown with superimposed red arrows). In each simulation, a physicomimetics-controlled hexagonal lattice (shown as black boxes marked with a white 'X') starts out in the top left corner, and locates the plume using a sophisticated casting algorithm. Once the plume is located, the lattice executes a given CPT algorithm for 1200 time steps, or in the case of a fluxotaxis-driven lattice, until the emitter is found. The lattice reverts to its casting behavior when it loses the plume. (Simulation videos were sped up to reduce download size.)

Anemotaxis

Anemotaxis-driven lattice fails to locate the emitter (video) because the instantaneous wind direction does not point to the chemical emitter, but rather to the current wind source. Due to its mass inertia, the shape of the plume depends on the history of the wind over a period of time, and as this experiment shows, instantaneous fluid velocity is not a reliable indicator of an emitter's location. Also note the characteristic behavior of the anemotaxis lattice once it moves upstream, past the emitter; it is now very difficult for the lattice to successfully locate the emitter, since whenever it detects the plume, it moves upstream and away from the emitter!

Chemotaxis

Chemotaxis-driven lattice fails to locate the emitter (video) because the plume consists of many small filaments, each of varying density, and chemotaxis always follows the chemical gradient, which causes the lattice to find pockets of high local chemical concentration, as opposed to finding the true chemical emitter. This simulation shows that local chemical gradient is misleading in turbulent plumes.

Fluxotaxis

Fluxotaxis-driven lattice successfully locates the emitter (video) because it looks for locations of maximum divergence of mass flux -- a second-order measure that avoids the problems encountered by the first-order schemes, such as anemo- and chemotaxis. Because of its theoretical foundation in fluid flow physics, fluxotaxis is able to determine when it located the emitter, and the lattice can terminate plume tracing early. This is not the case with other CPT approaches, where termination criteria are usually based on heuristics.

Encouraged by the good performance of our CPT approach in simulation, and with the support of the National Science Foundation, we have successfully implementing our solution using a laboratory prototype system. Using ethanol vapor as the trace element, we have collected extensive chemical plume data in a controlled flume (test chamber) environment. Furthermore, we have demonstrated repeated successful trials of CPT using three robots and our novel trilateration localization technique. Our most recent success is described in the SAB'06 paper, below. For videos showing our Maxelbot robots performing a wide variety of tasks, including chemotaxis CPT, see the Maxelbot robots in action!

Publications

Spears, W., J. Hamann, P. Maxim, T. Kunkel, R. Heil, D. Zarzhitsky, D. Spears, and C. Karlsson. Where Are You? Proceedings of the Second Swarm Robotics Workshop (SAB'06), 2006.

Zarzhitsky, D., D. Spears, and W. Spears. Distributed Robotics Approach to Chemical Plume Tracing. Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS'05), pp. 4034-4039, IEEE Press, 2005.

Spears, D., D. Zarzhitsky, and D. Thayer. Multi-robot chemical plume tracing. Proceedings of the 2005 International Workshop on Multi-Robot Systems., pp. 211-222, Verlag, 2005.

Zarzhitsky, D. and D. Spears. Swarm Approach to Chemical Source Localization. Proceedings of the IEEE International Conference on Systems, Man, and Cybernetics (SMC'05), pp. 1435-1440, IEEE Press, 2005.

Zarzhitsky, D., D. Spears, and W. Spears. Swarms for Chemical Plume Tracing. Proceedings of the IEEE Swarm Intelligence Symposium (SIS'05), pp. 249-256, IEEE Press, 2005.

Zarzhitsky, D., D. Spears, D. Thayer, and W. Spears. A Fluid Dynamics Approach to Multi-Robot Chemical Plume Tracing. Proceedings of the Third International Joint Conference on Autonomous Agents and Multi Agent Systems (AAMAS'04), pp. 1476-1477, IEEE Press, 2004.