Tested reactive resurfacing behavior. When the robot detects any type of fault (like low battery, high temperature, process crash, or unexpected obstacle) is aborts its current plan and returns to the surface. In the simplest implementation, the robot runs its vertical thrusters and heads straight up. Very often this works fine, but in an environment like la Pilita which is shaped like a vase, it can run into the bottom of an overhanging ledge. It might still bounce around enough to eventually pop out through the opening, but maybe not. We have been testing a variety of algorithms for reactively moving away from walls or towards free space while also moving upwards. This behavior is reactive because it does not rely on any prior information or even knowledge of where the robot is located. Getting this to work reliably has involved deciding how far to stay away from walls, how rapidly (or forcefully) to move away, and what rate to ascend. After trying several schemes we settled upon a method based on repelling away from walls at rate inversely proportional to distance. We then set about testing this by getting progressively deeper and farther underneath the ledge. Eventually the robot was returning on its own from depths of 50m while more than 15m back underneath a ledge.
Execute multigoal dive with contingency plan. Continuing on work from Wednesday we executed several 30 minute dives that involved navigating down to deeper areas where we have sparse sonar data and scanning intensely to try to collect enough readings to determine whether there is a tunnel present. These multigoal plans take the robot to the depths of la Pilita and must return on schedule otherwise the contingency plan is invoked and the robot begins to return home automatically.
Tune proximity operations parameters. We continued refining the proximity operations behaviors and in particular worked on enabling the robot to move smoothly along walls while maintaining a constant standoff distance. It is now able to stationkeep quite steadily but motion relative to the wall can become erratic for very rough or fractured walls. We also observed that the robot sometimes drifts upward along an inclined wall. Work will continue on this after rethinking the current formulation.
Exercise science payload. We performed our first sampling operations at depth just at dusk today. While at 50m and back along a wall beneath a 15m ledge, we directly commanded deployment of the sampling arm and pressed the vehicle forward against the wall to fire the coring mechanism. We also drew in water samples of the material that was stirred up, and of course took images. The core came back with a bit of wall material although not a complete core. The next step will be to integrate some of these science activities along with proximity operations into a longer multigoal plan.
Dave Wettergreen
Carnegie Mellon University Robotics Institute
Above: Probe Arm - Solid Sample Collection (you can also watch it here).
Above: This small pile of soft sediment was collected at a depth of 50 meters in the cenote La Pilita 40 meters back under the cave edge. The sample was collected by the solid core sample device on the end of the sample arm.
Above: Biogenic mineral crystal from the walls of La Pilita fell into the frame of the DEPTHX robot during a deep mission. These crystals are thought to be forming through biologic mediated processes.
Above: A top view of the DEPTHX robot with the sampling arm extending to test the operation of the solid core sample device. The coring devise is the gray cylinder at the front of the probe.
Above: A four-inch purple scorpion was found on the trail walking back from La Pilita. These scorpions are common in the caves nearby.
Above: Image taken with the stage 1 camera on the DEPTHX robot of the wall 50 meters deep in the cenote La Pilita. The texture of the wall is likely due to biogenic mineralization of crystals.
February 9, 2007
Agenda
Refine sample coring strategy
Develop complex contingency plans
Run multi-hour mission
Operate without tether
Status and Progress
Refine sample coring strategy. The rover’s sample coring mechanism can be triggered by a mechanical switch or by a signal generated by software. Our initial strategy was to station keep, while the arm extends, contacts the wall, and mechanically triggers the coring mechanism. The success rate has been low due to the difficulty in generating sufficient force on the trigger. Instead the rover was made to extend the arm and then move forward until motion stopped and the corer triggered and then the rover backed off retracting the arm. This method succeeded but suffered from occasionally bouncing off the wall before the coring mechanism fired. We are going to work alternatives.
Tested water sampling plumbing. The rover can sample water from its environment and we are particularly interested in do this during the coring operation when material is dislodged from the wall of the cenote. After simplifying the plumbing somewhat and inserting operations into the sampling plan the rover now draws water samples while in proximity to the wall. It carries 5 1-liter sample bags and we verified that it was filling the proper bag. During the DEPTHX science investigation these will be sterile bags and we will have to go through a careful procedure to sterilized and flush all of the plumbing through the pump and manifold that directs water to the desired bag.
Develop complex contingency plans. With the contingency plan based on moving upwards while repelling away from the walls working reliably, we began testing multiple stage contingencies. In this case the rover first tries to move back under its descent point and then ascends upwards to the surface. If any faults occur during the dive the rover begins execution of this plan, further faults associated with approaching walls or position estimation errors, cause the system to fall back to the purely reactive wall-avoidance behavior. The multiple fall-back contingency plan executed as expected.
Operate without tether. With a plan to dive, fly a simple pattern, and return to the surface loaded and multiple contingency plans queued up, we removed the fiber optic tether from the rover. The tether had been used thus far to monitor onboard software in real-time, so this dive the robot was on its own. After 21 minutes and 5 seconds the rover resurfaced within a meter of its descent point having completed the mission. We loaded a second plan, running circuits down to 60 meters. In this dive the rover got within 10 meters of a wall (while at a depth of 50 meters) and triggered a proximity fault. While it didn’t complete the dive plan completely, it did successfully execute its contingency plan and return to the surface (in 00:18:55). After adjusting the plan and bounds on the contingency fault, the dive plan was repeated and completed successfully (in 00:33:10).
Run multi-hour mission. We then programmed a longer mission, of vertical “boxes” pivoted off center to create a star pattern that would provide very dense observation of the lower chamber of the cenote. The plan was 2280m in length. The plan was given a timeout at 10800 seconds (3 hours). The vehicle dove at 10:24pm and resurfaced again at 1:34am about 1.5m from its descent point. The complex plan had required the rover to reach many waypoints and subsequent analysis revealed that it has spent much time making fine adjustments to reach the goals within 10cm. As a result its top speed of 0.2m/s was reduced to an average speed of 0.1m/s and it had timed out on the mission after 3 hours. Again the contingency plan functioned correctly and brought the rover back to the surface.
Dave Wettergreen
Carnegie-Mellon University
February 10, 2007
Agenda
Test altimeter-based coring
Run map-based localization
Status and Progress
Fire sample corer on altimeter. The rover’s sampling arm has an altimeter that was intended to allow it to adjust its distance to the wall the ensure proper focus of the close-up camera. We tied the altimeter to the coring operation so that when the range to the wall stabilizes to a constant (small) value the coring mechanism fires via a software signal. The proved more reliable than mechanical trigger which requires a smooth flat wall surface.
Improved wall approach. We also worked to continue refinement of the wall approach proximity operation. In final approach the rover only uses its rear thrusters for more stable control. It also moves with constant steady thrust rather than thrusting towards the wall and then coasting in. This has reduced the bounce, rotation or both that can occur when contacting and irregular wall, which seems to be most surfaces in la Pilita.
Ran map based localization. The dead reckoning algorithm, which uses an inertial measurement unit (IMU), depth sensors, and a Doppler velocity logger (DVL) along with a vehicle motion model has performed remarkably well. It’s drift is well under a meter and hour as long as all sensors remain locked. In particular a Kalman filter is able to maintain position across dropouts of the IMU and DVL but errors do accumulate. Using sonar-derived maps of prior and current dives, the robot can recognize its position and correct for any drift. (More specifically, the rover maintains a large number of possible maps and locations and determines based on probabilities its current best guess.) We ran a two hour dive to complete the mission of the Friday night and used the data previously collected to provide a map of the cenote. While the rover traveled it estimated its position from its sensors and also from the sonars and prior map. It completed mission in 01:43:00 (untethered). It’s dead reckoned estimate was off by about 1 meter, but its map-based localization was correct to 15 centimeters. This gives us some confidence that the rover will be able to complete dives in Zacatón that will require 6 hours or more and still reach goal and return to its starting point.
Dove under a dome. In a last couple dives before wrapping up we ran a couple short missions into areas of the cenote that had been sparsely observed. In a 22 minute autonomous dive, the rover dropped down to 30 meters traversed, literally underneath our camp, and rose up into a dome in the ceiling of the cenote. It then dove down and returned to the opening. In a similar operation the rover dove to what we thought might be a tunnel opening at 55 meters. It collected more data and narrowed the area where a tunnel might exist but left the question of an underwater tunnel unanswered.
