Carp Tracking

From RSN
Jump to: navigation, search

The common carp is an invasive species of fish which poses a significant threat across the Midwest. This species pollutes lakes by uprooting plants and releasing large quantities of harmful nutrients while bottom-feeding. It is important to track and control the species. Sorensen lab at the University of Minnesota has discovered that when carp form huge shoals they can be netted and removed with very high efficiency. However, finding and controlling these aggregations from manually using conventional boats is expensive and nearly impossible.

We lead a collaborative project funded by the National Science Foundation to build a network of robotic devices which can be used for locating and tracking radio-tagged carp. The project brings together roboticists, computer systems experts, mathematicians as well as fish biologists to tackle many challenging research problems ranging from optimal search and active tracking with multiple robots to energy-efficient operation as well as understanding fish behavior. Stay tuned for more on our results! In the mean time, you can check out the videos and publications on this page or contact Prof. Volkan Isler for more information.



Husky with radio equipment Lavant
Our system is built on a combination of two commercial platforms: the six-wheel differential drive Husky A100 from Clearpath Robotics for winter operation, and robotic boats used in winter. The overall system comprises of:
  • EEE PC running Ubuntu with ROS
  • Garmin GPS 18x for robust waypoint navigation
  • Radio antenna mounted on pan-tilt unit
  • Radio telemetry equipment from ATS-Track.

It is designed to search a lake until it detects a nearby fish then localize the fish using bearing measurements. We are also addressing the problem of long-term autonomy by incorporating solar-power-aware path planning. We address several problems, which include the design of a robust mobile sensor network, Optimal Coverage, Adaptive (or "active") Tracking, and Optimal Search.

Solar Power

More information can be found on the Solar Power page.

Our carp tracking robots usually deplete their battery in just a couple hours at most, so to achieve long term autonomy the addition of solar panels is necessary. These solar panels provide some interesting path planning problems. Fundamentally we would like to travel in the sun rather than the shade, if possible, but if the shortest path is in the shade it might not be worthwhile to deviate. We need an accurate solar map of the environment, as well as an accurate sense of the energy required to drive along a given path.

We use two different methods to find the solar map, both using no additional sensors beyond GPS localization and a current sensor on the solar array. We use Gaussian Process regression to construct a solar map for an arbitrary environment; this method works well for a short time period but it doesn't have a good way to account for the movement of the sun. On a lake we can use raytracing, taking advantage of the fact that obstacles are likely to appear on the shore, to determine the positions of these obstacles. This gives us a time-varying map.

Optimal Coverage

Carp behavior has been heavily studied in an effort to develop population-control strategies. We can use prior information about likely carp locations to guide the robots as they search a lake. We incorporate this information as convex regions which the robot must cover and propose an algorithm to design the robot's path through and between these regions. This problem is a generalization of the well-known Travelling Salesperson Problem. Since it is known to be NP-Hard, we designed a constant-factor approximation algorithm. For more information, please refer to [7]

The videos above are from field tests in Lake Keller and Lake Staring, in Minnesota.

Optimal Search

Recently, we have begun studying the problem of searching for a moving, or "wandering" fish. In this case, we model the fish movement as a random walk and solve for the optimal tradeoff between search and solar energy harvesting. For more information see Related Publications.

Active Localization

More information can be found on the Active Localization page.

The radio antenna provides bearing estimates from the robot towards the tagged fish. However, the bearing estimates are not precise: Even after 3-4 bearings measurements are collected and combined, the target uncertainty may still be large.

The goal of active-localization is to select sensing locations for the robot so as to minimize the uncertainty in the location of the target in as few measurements as possible. For this purpose, we proposed active localization algorithms and evaluated them. We provide closed-form guarantees of the final uncertainty as well as the time required to localize a target. In particular, we analyze the trade-off in time-to-localize versus the accuracy of the final estimate. The robotic system uses these algorithms to localize targets accurately.

Related Publications



This project has been featured in print and televised news. See the following for more information.


This work is supported by the National Science Foundation: RI: Large: Collaborative Research: A Robotic Network for Locating and Removing Invasive Carp from Inland Lakes

Our preliminary work was supported by NSF Projects 0917676,0907658 and 0936710, a fellowship from the Institute on the Environment at the University of Minnesota, and McKnight Land-Grant Professorship.

Personal tools