The Boston Aerial Robotic Team Helicopter

The Boston Aerial Robotic Team (BART) was led and coordinated by Draper engineer Paul DeBitetto, the vision system was design and implemented by MDSP Lab [ http://acadia.bu.edu/ ] member Michael Bosse [ http://acadia.bu.edu/zanj_html/zanj.html ], the simulation suite was written by Draper Lab [ http://www.draper.com/ ] employee Eric Johnson, the tracking and navigation subsystem was designed and built by Draper engineers Eric Johnson and Ed Lanzilotta and MIT [ http://web.mit.edu/ ] student William Hall.

MDSP Vision System

A key element of the robot helicopter was a vision system designed by Mike Bosse [ http://acadia.bu.edu/zanj_html/zanj.html ] that used real-time video imagery to detect and classify the toxic waste barrels during the flight.

Block Diagram

The vision subsystem receives the video stream sent via a 2.4 GHz RF signal from the camera onboard the helicopter. Each frame that is processed is buffered until the corresponding helicopter state information (roll, pitch, yaw, x, y, z, time) comes via ethernet from the ground control computer. The state information is needed to estimate the expected size of the labels in the image, and to transform the location of a drum in the image into field coordinates. The location and classification of the drums are then sent to an optimal estimator which determines the best positions of the drums in a series of frames.

Typical Scene:

The first block is a synthesized scene for a typical view from the camera in flight at about 10 ft over the drums. The next block shows the gradient image, using magnitude of Sobel derivatives. The labels on the drums characteristicly have a high concentration of edges surrounded by a relatively smooth rim. This feature is picked out by a matched filter, lower left block. The pixels around the peaks of the output from the matched filter are then processed to determine label type.

In the future the team would like to use the images from the onboard camera as a third position measurement sensor and as a terrain mapping sensor. This would allow the vehicle to fly and / or land in areas of poor GPS coverage.

Guidance and Navigation

The sensor suite of the helicoptor included:

Novatel RT-20 differential GPS
Six-DOF inertial measurement unit
Downward-looking video camera
Compass (2-axis magnetometer)
Ultrasonic altimeter sonar

The computer controller and data processor was a PC-104 format 486 running at 50MHz. It drove the helicopter's servos through a modified radio-control receiver.