Shark Aggregation Tracking

Starting Fall 2015, LAIR is collaborating with biologists from the University of Saint Katherine and CSU Long Beach's Shark Lab to model and track animal aggregations using AUVs. Previously, LAIR had developed motion planning algorithms that enable multiple AUVs to track a single shark. This research extends previous work and aims to provide a novel method for tracking entire aggregations of animals using position estimates of only a subset of individuals within an aggregation.

Problem Definition

The goal of this project is to estimate the 2D planar size, location, orientation, and number of individuals of an animal aggregation using robots equipped with sensors that can measure the position states of a subset of individuals within the aggregation. We introduce a pair of coordinate axes based on the attraction line, with ρ as the distance from the line and ψ as the distance along the line from the line’s center.

Overall System Block Diagram

The proposed system is summarized in the figure above. A historical data set is first processed to determine the parameters of a stable multi-agent swarm model. In this case, we use a swarm model based on Artificial Potential Fields (APFs), where the model parameters are attractive and repulsive gains.

A graphical representation of the artificial potential field model.

The swarm model parameters are used in Transition Matrix Modeling to run a large number of simulations in which aggregation parameters (e.g. n, L) are varied. From these simulations, Markov Matrix model parameters ρ90 and ψ90 are determined, which correspond respectively to the distance from the line, and the distance from the center along the line within which 90% of individuals in the aggregation are found.

Tagging a Leopard shark with our SmartTag

During real time aggregation tracking with an AUV, new sensor measurements of tagged individuals are used by Individual Particle Filters to estimate the state of each tagged individual. The estimated position states of the tagged individuals are used to estimate the endpoints of the attraction line using the Attraction Line Particle Filter. The estimated position states of the tagged individuals and the estimated line state are used to estimate the aggregation state (e.g. number of individuals in the aggregation, length of attraction line) using the Aggregation Particle Filter.

Shark Swarm at La Jolla, San Diego

The system has been validated with an aerial video data set of a leopard shark aggregation. For example, when 40% of sharks in an aggregation are tagged, the estimated aggregation size (i.e. total number of sharks) has an error of less than 27%.

Here are some of our published work and presentations on this project:

  • Predicting Coordinated Group Movements of Sharks with Limited Observations using AUVs, Ho, C. , Joly K., Nosal, A.P., Lowe, C.G., Clark, C.M., To appear in ACM Symposium on Applied Computing (SAC 2017), Apr., 2017. [Preprint]
  • “Predicting Coordinated Group Movements of Sharks with Limited Observations using AUVs.”Southern California Conference for Undergraduate Research (SCCUR), November 2016. [Link]

Shipwreck Mapping

Starting Summer 2016, the LAIR has been collaborating with Cal Poly SLO and archaeologist Dr. Timmy Gambin from Malta to intelligently search and map shipwrecks using an AUV. This includes detecting shipwrecks using image processing techniques on side scan sonar data and motion-planning to map the shipwrecks. In Summer 2016, the team succesfully deployed the AUV at two archaeological sites off the coast of Malta. This project is funded through an NSF IRES grant. More information about the project can be found at the Malta Shipwreck Mapping Site.

A video accompanying the 2017 ICRA paper of this research:

A video accompanying the 2018 ICRA paper of this research:

Malta Cistern Mapping

In the spring of 2008, Dr. Clark worked with archeologist Dr. Timmy Gambin from Malta to explore and map ancient cisterns located on the islands of Malta and Gozo. The cisterns of interest acted as water storage systems for fortresses, private homes, and churches. They often consisted of several connected chambers, still containing water. A sonar-equipped Remotely Operated Vehicle (ROV) was deployed into the cisterns dating back to 300 B.C. to obtain both video footage and sonar range measurements. Different mapping and localization techniques were employed to construct 2D maps of 6 different cisterns. An additional expedition took place in Spring of 2009, resulting in 26 more sites being mapped. This project was initially funded through a California State Faculty Support Grant, and is now being funded through an NSF IRES grant.

More information about the project can be found at the Malta Cistern Mapping Site.

A 2D map of a cistern