|February 5, 2007
FIRST DATA: Today was a milestone for Team DEPTHX. The vehicle made its first true exploration run at Cenote La Pilita (that is, it built a 3D map of terrain never before seen by man). Mission control was established in two large tents beside the 25 m diameter cenote. Marcus Gary and John Kerr managed the “sky trak” mobile hoist to get the vehicle in the cenote whereupon John spent a fair bit of time in the 32C water performing fine tuning of the vehicle buoyancy by adding small lead drop weights. For these initial missions, largely because of the unknown spatial extents of La Pilita, we equipped the vehicle with a 2.5mm diameter Dyneema safety tether (breaking strength 1.5 metric tons) and a 1 mm diameter fiber optic link for observing the real-time behavior of the vehicle at mission control.
Importantly, besides providing a huge stream of state and sensor data (see previous blog image of the microRAPTOR real-time vehicle state software output) we were able to monitor live what the vehicle was “seeing” through an openGL visualizer that showed a representation of the vehicle along with sonar beam models for the 54 discrete sonar transducers that build the world map about DEPTHX. The present onboard software can reject spurious readings in real time -- important for the upcoming geometry-based navigation tests. Also piped up are video images from the three onboard cameras. An example of the output from the wide field science payload camera is shown below in the screen capture from one of the mission control laptops.
Left: DEPTHX being lowered into the 32C hydrothermal spring of cenote La Pilita.
Above: Dom Jonak (left) and Nathaniel Fairfield at mission control, La Pilita monitoring the live data uplink from DEPTHX. Current testing is being done on-tether for real-time performance evaluation. Below: John Kerr releases the vehicle to begin testing.
Following a series of warmup runs the vehicle was commanded to undertake a vertical powered descent mission to a maximum depth of 100 m while insuring that it maintained a “keep out” zone of 10 m about the vehicle and a minimum 5 m standoff distance to bottom. The resulting map (see figure below) shows 340,000 sonar wall hits during the course of an approximately 1 hour duration mission. During the descent the vehicle was set in a uniform rotation so that the sonar arrays “scanned” the wall, creating a high density fill factor.
Left: George Kantor (left) and Dave Wettergreen (both from Carnegie-Mellon University) contemplate the performance of DEPTHX at cenote La Pilita.
The real power of the 4-Pi steradian imaging system is that DEPTHX can simultaneously look in all directions. This, and the fact that the vehicle is descending while it rotates, provides the maximum opportunity for filling in voids that would otherwise not be seen by normal planar line-of-sight imaging systems. This capability is crucial to the implementation of 3D SLAM, which we will begin testing towards the end of this week.
Of particular interest, the plan view of the map (left image below) shows the presence of a 20 m diameter tunnel leading off from the western wall of the chamber at a depth of approximately 40 m at the roof and leading in a northwest heading. Similarly, there appears to be a broad bulging in the room in the northeast corner at about the 30m depth level. The vehicle came closest to the floor on an eastward sloping incline; what happens to the west (which is in the direction of cenote Zacaton - the target of the DEPTHX finale field exercise in May) is unknown. We hope to fill in more data in the coming days leading to an untethered subsurface mission in a full 3D labryinthine environment.
ABOVE: DEPTHX 3D map of Cenote La Pilita, created February 5, 2007 during the first true exploration mission for the DEPTHX project. The map is comprised of 340,000 individual sonar “hits”. In general the vast number of points clearly define an enormous subterranean submarine void - the full extent of which will not be determined until further off-axis exploration in the coming days. Outlier points are raw data as recorded by DEPTHX; some of these were a result of an incorrectly cabled sonar transducer; some represent “multipath” signal bounce (and therefore non-existent geometry); and some likely represent sparse new data suggesting potential voids to be explored on subsequent missions. (click here for a bigger version of this map) BELOW: screen capture at mission control showing wide field camera shot of the wall of La Pilita at 12m depth and live sonar imaging (right).
February 6, 2007
Science Payload Powers Up:
We continued with vertical powered exploration profiles today, reaching a maximum depth of -98 m in cenote La Pilita. During this time we also tested the Science Payload wide-field camera at several depths and got our first look at the bottom of La Pilita, where, in true planetary robotic tradition, features began to acquire names, such as the “Iguana Rock” shown below.
Above: The eye of the bot-wide field camera on the Science Payload for DEPTHX.
Above: Bio film on the wall of cenote La Pilita at a depth of -12m.
Above: “The Iguana” rock on the bottom of La Pilita at a depth of -105m.
Above: Nathaniel Fairfi eld gets to experience the world of the bot (at shallow depth).
Above: The busy world of Mission Control at cenote La Pilita.
This evening Alejandro Davila picked up SwRI researchers Tom Lyons and Ian Meinzen who will be with us for the remainder of Mission 1 to tend to hardware and software relating to the Science Payload. Below is appended Dave Wettergreen’s summary of the technical results from February 7.
Tecnical Status and Progress - February 6
- Tested proximity operations. The robot typically tries to stay away from the walls, but in order to take images and collect samples it must move into close proximity of the wall and then position steadily (stationkeeping) to move along a transect. Proximity operations is our term for this mode. Not surprisingly La Pilita presents a much more complex geometry than the smooth walls of a test tank. The algorithm fits a surface to a collection of forward sonar range measurements and then tries to control the vehicle’s thrusters to move to a point relative to the vehicle (for example its probe tip) along the wall. When the wall has concavities and convexities, the tracking point can jump around or not move at all as the vehicle moves. To correct for this, the vehicle might move erratically. This is not a desirable behavior so we modified the algorithm to instead drive the vehicle smoothly and then reacquire a point on the wall. This was tested in a number of locations in la Pilita and we are now able to get the vehicle into position for science investigations of the complex geometry of la Pilita’s walls.
- Collect sonar data. It is important to characterize the performance of the sonars on the robot. We weighted the vehicle down (making it negatively buoyant) and then hung it at 22m and 42m depth to collect sonar data. In particular we adjusted the gain on the sonars, the maximum range, and low pass filtering. By hanging statically in the water we were able to see when the sonars were returning consistent (and believable) ranges. We adjusted all the major parameters to a variety of settings, trying to bracket the best values. There is no one perfect setting because each transducer behaves slightly differentlywe have 32 of the 100m sonars and 24 of the 200m sonars so we were looking for a best overall performance. We think we’ve found that and furthermore have got settings that should work in Zacatón as well.
- Dropped to the bottom. With the sonars tuned we dropped the vehicle to the bottom of la Pilita, again on a tether for safety. We dropped in a different location and reached the bottom at 98.3m. This time we had added dive lights and so we took pictures of the bottom including a rock we called the iguana. The robot actually touched down gently. Using the live visual image (something we won’t have when we remove the fiber optic real-time data upload cable and go fully autonomous and un-tethered) we used the thrusters to hop around on the bottom. On this second drop to the bottom and return up we collected a second complete sonar data set of the cenote which we will use to build an even more accurate map.
Dave Wettergreen, Carnegie-Mellon University
February 7, 2007
Wall and Water Sampling + Pre-SLAM: Most of the daylight hours today were consumed with tests of the various Science Payload sub-systems, mainly pre-programmed capture of up to five independent 2 liter water samples and the test firing of the coring tool, which extracts a 1 cm diameter x 3 cm long bio sample from a candidate wall location - see the images below for one of many such test sequences conducted today. In these photos you can see the Science Payload sampling probe in its extended position. The purpose of this design was to permit the main DEPTHX vehicle to “stand off” from areas of potential biologic interest and send in a non-disturbing probe. Bringing the entire vehicle up against any surface in a 3D labryinthine environment is fraught with possibilities for entanglement (and therefore potential loss of the vehicle) so in general we maintain a significant proximity fault “keep out” zone of 5 m radius from the vehicle. Only during proxops (proximity operations) do we intentionally allow the vehicle to get within 1.5 m of the wall; the probe then goes the last distance to look at what’s on the wall. This also has the benefit of minimizing thruster disturbance of wall materials.
|Above: DEPTHX, with its Science Payload probe arm extended (left) moves in to acquire a biological wall sample (right)
Above: DEPTHX begins autonomous mission 7 after dark. The bot made a powered descent to 80m and successfully conducted exploration circuits at three depth levels.
Above: inside the mind of the bot -- a 3D slice of the “evidence grid” from -40 to -60 m in La Pilita.
Later in the evening we began a series of more ambitious autonomous missions in which we sent the bot on a pre-programmed trajectory to various depths at which point it would then execute a triangular-shaped maneuver - moving 15 m along each leg of the triangle and then returning to the original vertex. The bot would then move to a shallower depth level, about 15 m higher than the first, and duplicate the maneuver. After doing this at many levels it would return to its “home” position, which is shown in the above night shot of the bot with its twin HID head lights blazing. The first such mission was attempting to reach 60 m depth when an over-temperature fault indicator on the inertial guidance unit tripped and the vehicle fell back to an abort behavior - surfacing along a path that placed it at the geometric centroid of the cenote. From the surface, except for the data being displayed on the microRAPTOR computer console interface, there was no indication that anything was different until the HID lights began making the entire cenote glow from a depth of around 20 m. After consulting the IMU operations manual we raised the temperature fault trip point and sent the bot back down, this time to -80 m. It then successfully conducted two back-to-back autonomous missions, returning to “home” each time... the first with a navigation error of just 9 cm; the second with about 0.5 m accumulated drift following more than an hour of operation away from base.
On each of these missions DEPTHX continued to collect a fire hose stream of geometric data, bit by bit filling in the unknown voids in La Pilita. Using these data a 3D “evidence grid” -- a probabilistic density map showing “voxels” (cubes) of space likely to represent the internal boundary surface of the newly explored terrain -- was constructed. One of the interesting features of this approach is that “negative probability” is deposited in the grid in places where the sonars detect nothing to exist. This negative probability accumulates over time to suggest that there really is nothing there, a consensus of many different sonar transducers seeing the same area over time. And thus, one can make a reasonably safe assumption that in such areas it is OK to drive the bot. Coincidentally, the converse is also true: that the places where true boundary surfaces exist will become better defined over time and have a higher probability rating that they are real. These, in turn, form the basis for the creation of a map, from which the vehicle location can be deduced in 3D. This, in a nutshell, is SLAM -- Simultaneous Localization and Mapping. The green figure above shows the 3D evidence grid for La Pilita as it is being developed mission-by-mission. It is specifically a slice from -40 m through -60 m in depth. One can clearly see in this image that the core of the cenote is black, meaning high negative probability of occupancy of something real - -and it is within that zone that it is safe to navigate. With some luck we will conduct the first full 3D SLAM navigation by Saturday. The expedition packs up on Sunday, February 11.