Novus Light Home

Blog

Rendering of Gemini

The Cornell University Autonomous Underwater Vehicle team (CUAUV) is an undergraduate student project team of students that build robotic submarines to enter in the International RoboSub competition. CUAUV has placed in the top ten every year that it has entered and has won the overall competition six times, most recently in 2014 with its vehicle Gemini.

RoboSub competition tasks range from shape and color recognition to torpedo firing and fine manipulation of small objects. The team is continually refining Gemini, and its current iteration, called Argo, features stronger frame structures and pressure vessels, a more streamlined electronics system, an improved sensor suite and more robust vision algorithms. Gemini’s sensor suit is designed for use in navigation and data collection. It includes machine vision cameras, compasses, inertial measurement units (IMUs), a Teledyne RDI Explorer Doppler velocity log, a depth sensor, an internal pressure sensor, and a hydrophone array.

The Gemini UAV being tested in Cornell's indoor pool

Gemini being tested in the Cornell indoor pool.

To autonomously navigate through the competition course, the vehicle is equipped with two main classes of sensors: one to observe the vehicle’s environment, and one to determine the vehicle’s state. Visual recognition and navigation tasks are performed with two forward cameras and one downward camera, while the recovery task is handled by the hydrophones system using a passive acoustic array.

The Teledyne RDI Explorer Doppler velocity log (DVL) provides accurate three-dimensional velocity data. This information is used in conjunction with the other sensors to provide closed-loop vehicle control.

The vision-processing system is designed to give the vehicle up-to-date and accurate data about the surrounding mission elements. A vision daemon handles, reads, and processes camera data. Prior to processing, raw camera data is passed through an undistortion filter. The undistorted data are analyzed using a combination of color thresholding, Canny edge detection, contour analysis, and Hu moment characterization. Gemini uses two forward facing Allied Vision Guppy F-080C cameras with wide angle lenses that enable a stereovision system to determine distances to objects, which is currently a work in progress. According to Mark Lee, team co-leader, they have successfully implemented a rudimentary version of stereo vision based on the difference in the position of an object seen in two viewpoints with a single camera. “We were unable, however, to successfully implement true stereo vision with both cameras; issues with extrinsic parameter calibration and tuning stereo algorithms to produce an accurate depth map proved very challenging to overcome,” Lee explained.

With the 2015 vehicle, Argo, the team made an active decision to forego further work on stereo vision to achieve greater advancements in the mechanical design of the AUV. Lee said that the team will add imaging sonar on the new vehicle, which will enable the team to transition from depending on stereovision to complete variety of submarine tasks such as localization and path planning.

Vision processing

Each mission element has its own vision module that can be started independently of every other module. This lets the team reduce processor usage by only running necessary processes and by allowing for multiple modules to be run in parallel. Mission elements are pinpointed by a combination of edge detection, color thresholding, and contour analysis, though not all elements use every algorithm.

Automated vision

An automated vision evaluator called Cave (CUAUV Automated Vision Evaluator) helps analyze vision performance by keeping a database logged video and providing graphical framework for quick annotation and automated testing. CAVE organizes captured logs and allows searching by metadata, including information such as the weather conditions, location, and mission elements present in a video. Additionally, the system can play video files on demand and stream frames directly to the existing vision framework.

With new mechanical features, revamped electrical systems, and more robust mission planning software, Argo will be an exciting advancement for CUAUV.

A 3D vehicle simulator built on the open source Panda3D simulation engine is used to verify mission and vision code before it is brought to the pool. The simulator can use the same code as the vehicle and saves many hours of in-water testing, providing visual feedback to the software team during development.

Challenges and success

The team has 15 years of aggregate research and development to enhance AUV technology, and Argo is a culmination of that work. It has not been without its challenges, however. “The rigorous design and manufacturing schedule to build a new AUV in 10 months can be very challenging, especially as college students balancing coursework and various commitments,” Lee noted. He added that an additional challenge is adapting to the outdoor competition course in San Diego from the indoor university pool, in which the latest Gemini has been fine tuned.

The next step for the team is manufacturing the 2015 vehicle, Argo. Lee said, “With new mechanical features, revamped electrical systems, and more robust mission planning software, Argo will be an exciting advancement for CUAUV. “

Written by Anne Fischer, Managing Editor, Novus Light Technologies Today

Labels: RoboSub,AUV,vision,imaging,Cornell,Allied Vision,Teledyne,Underwater Vehicle,submarine

Back Back to Enlightening Applications
 

Illuminating Products

Copyright © 2017 Novus Media Today Group, LLC. All rights reserved. Website design and build by MM Design.