We ran our first glacier mission today. We want to use sonar data to develop a profile of Taylor Glacier where it descends into Lake Bonney and to build a mosaic of glacier face images taken by the forward-looking camera.
Picture of the glacier. We want to profile part of this glacier that is underwater. Thanks to Vickie for this picture.
Today’s mission was mostly exploratory – we had absolutely no idea what we would find. Many icebergs were visible on the surface and we were wary of running the robot into them while appproaching the glacier. We approached the glacier slowly while stopping at places to rotate the vehicle to do a full scan. The robot has a multi-beam sonar, Delta T mounted at its front. The Delta T consists of 480 sonar beams forming a two dimensional “fan”. Each ping consists of the range returns and intensities of 480 beams. The ping rate is 3-4 Hz. Thus, the Delta T returns can be used to visualize a wall or an obstacle in front of the robot. We can rotate the robot to get a 3D view of the surroundings. The Delta T visualizer along with a simulator developed by Shriram Ayer with Andy Johnson at the Electronic Visualization Laboratory form excellent visualization tools that help us in decision making.
We had plenty of excitement today. Bill and Vickie were out tracking the robot with a radio beacon and flagging specific locations to obtain their GPS coordinates. The ice near the glacier had gotten thin and uneven due to the warming effect of the waterfall created by glacier meltwater. A wheel of their ATV went into an ice ditch, the ATV rolled over and Bill’s leg got stuck under it. The ditch had a thin layer of ice on it with water underneath it and Bill’s leg got soaked. Bill had to come back to the Bot-House to change into dry clothes.
We did not find any icebergs under-water, but we found that the glacier face had a cleft with a cavernous region beneath. Below is a screenshot of an image from the Delta T visualizer.
Glacier face as seen in the Delta T visualizer. Thanks to Chris for this screenshot.
We drove the vehicle close to the glacier face in the expectation of getting good images, and we ended up driving the vehicle into the cleft. The depth of water below the vehicle was about one meter and this led to loss of DVL lock. DVL (Doppler Velocity Log) is one of our primary navigation sensors and is crucial in estimating the position of the robot. If we lose DVL lock, the position estimate quickly drifts and the vehicle loses track of its correct position. We had discussed this situation before the mission and took swift action. We switched the mode to joystick control and Chris used the joystick to drive the vehicle back off the ledge. The position estimate had drifted by about 20 meters during that time. We then drove the robot back to a known location that Bill and Vickie had previously marked. We used the GPS coordinates of this location to reset the robot’s pose estimate and the robot was able to drive back to within 2 meters of the melt-hole.
Below is a snapshot of the visualizer taken when the vehicle was moving away from the glacier face.
Screenshot of the run-time visualizer. The robot is in the centre of the screen with sonars shooting off of it. The glacier face is seen as the wall of green dots to the left. The line of yellow dots is the lake ice. The circles are locations where the robot rotated to perform a scan.
Below is a picture of the glacier face taken by the forward looking camera when the robot was very close to the nose before getting into the cleft.
Photo of the glacier face taken by the forward looking camera. Thanks to Kristof for this image.
I extracted the DeltaT data and plotted it up after some cleaning up.
Plot of the Delta T data. The lake bottom is seen curving into the glacier face at the right.