Collaborative Research-Remote Imaging of Community Ecology via Animal-borne Wireless Networks

This project features conception, design, and deployment of a wireless network of embedded devices, for monitoring the behavior of animals in the wild.  The system is being deployed and tested in biologically relevant scenarios. This project has been the first to deploy animal-borne wirelessly networked devices that are capable of providing not only geo-location data, but also of executing cooperative strategies that save battery-life by selectively recording bandwidth-intensive audio and high-definition video footage of occurrences of animal group behavior of interest, such as predation. In addition to enabling autonomous video capture, the wireless network registers the relative positions of the animals and other sensory information that will be useful in sociobiological characterizations. This project is addressing three primary technical challenges:

Investigating methods to design and analyze the performance of distributed algorithms that implement autonomous decisions (for video capture) at the mobile agents, subject to communication and computational constraints and costs.  

We have proposed a new class of distributed state estimation algorithms, and characterized networks properties under which asymptotic state omniscience can be achieved. We have also shown that ours is the most general class of algorithms for which asymptotic state omniscience can be achieved. Our algorithms have been implemented in our systems to facilitate event-driven distributed decision-making.

We also investigated the design of a remote state estimation system for a self-propelled particle (SPP). Our framework consists of a sensing unit that accesses the full state of the SPP and an estimator that is remotely located from the sensing unit. The sensing unit must pay a cost when it decides to transmit information on the state of the SPP to the estimator; and the estimator computes the best estimate of the state of the SPP based on received information. In this paper, we provide methods to design transmission policies and estimation rules for the sensing unit and estimator, respectively, that are optimal for a given cost functional combining state estimation distortion and communication costs. We consider two notions of optimality: joint optimality and person-by-person optimality. Our main results establish the existence of a jointly optimal solution and describe an iterative procedure to find a person-by-person optimal solution. In addition, we explain how the remote estimation scheme can be applied to tracking of animal movements over a costly communication link. We also provide experimental results to show the effectiveness of the scheme.

We are pursuing data-driven fundamental research on the modeling of animal group motion for multiple sociobiological configurations that will promote a formal understanding of the mechanisms of social interaction.  

We have developed an analytically tractable model of pursuit and evasion in the plane in order to study individual evasive strategies and group-level dynamics for a heterogeneous herd under pursuit from a single predator. Heterogeneity in individual maximum speeds is used to represent variation in age and ability among herd members. We show that a pursuer strategy of optimal target selection guarantees capture of an individual in bounded time for both global and local sensing regimes. We propose evasion strategies and prove conditions under which they guarantee capture avoidance. Properties of networks to identify leaders and collective decision-making principles have been identified.

Driven by analysis of video data of zebra herds evading a remotely controlled predator, we have extended our analytically tractable model of pursuit and evasion.  We account for key constraints on speed, lateral acceleration, and turning and leverage differential game theory to derive optimal pursuit and evasion strategies.  We show that an evader’s domain of danger depends both on relative positions and relative headings among evaders.  Model predictions of collective dynamics inform the design of algorithms for efficient use of crittercams.

This project has a significant component of applied research on methods for hardware integration for building distributed networks of embedded devices that are capable of executing our newly developed algorithms, subject to power and weight constraints. 

In collaboration with Gorongoza National Park (Mozambique), we have performed our second deployment of 13 units of last-generation devices on two animal species. The rich variety and volume of collected data will enable the investigation of unique animal group behavior as various species return to their habitat after a long civil war. We will also perform data-driven performance validation and optimization of our system for future deployments. 

Alumni:

•    Dr. Shinkyu Park (UMD) has defended his Ph.D. thesis and is now a postdoctoral researcher at MIT (Senseable City Lab / CSAIL)

•    Dr. Konrad Achenbach (NGS/UMD) has completed his postdoctoral work and is now working for Proteus Digital Health, Inc

•    Mr. Manjur Ahmed (UMD) completed his undergraduate degree and is now working for Accenture.