2009-11-13 Mission Success

We came, we saw, we planned a path. The gliders are still swimming, though it's suspected that RU05 has a broken wing. The same storm has claimed the lives of several New Jersey fishermen and caused nautical mayhem up and down the East coast.

Today, the Rutgers pros will be taking control of the gliders again, guiding them safely back into port around shipping lanes.

2009-11-12 Noreaster

For the past 36 hours we've been battered by a new weather system, a classic Noreaster that wouldn't appear in yesterday's current predictions. Here's the CODAR from yesterday as the storm moved in:

• 700 GMT MARC_oi1h_2009_11_11_0700.jpg
• 1500 GMT [MARC_oi1h_2009_11_11_1500.jpg

The gliders are barely holding position, or being pushed to the South West. They spend a lot of time at the surface, and in just a few minutes the tail in 40-knot wind can catch enough of the breeze to erase 3 hours of swimming upstream. Future generations of path planners should take this into account, that the glider is essentially a two-state machine. While on the surface, it acts like a drifter!

More feature requests for future mission planners:

• perform a translation between the desired glider behavior, which should be valid & achievable, with the waypoints passed to the gliders, which might be "faked."
• Fine-tunability of surfacing times to control tradeoffs between controllability, battery power, data transfer, etc.

Our two main goals, supplied by Oscar, are to keep the gliders mid-shelf and in deep water (i.e. off the 30m isobath). At the same time we want to protect RU23 from falling off the shelf. To this end we'll bring it directly into shore, while the others try to stay stationary by fighting more or less directly against the current. David Aragon provided an elegant metaphor to describe the glider's behavior in these strong currents: it's like a floating object tied to a string and immersed in a current: it will eventually swing around so that the current is pushing it directly away from its goal.

Note that the path planner requires waypoints to be a part of an executable, valid plan (i.e. it must believe you can actually reach the waypoints in the required time). No plan returned a valid path to the desired locations, they all defaulted to the standard "linear interpolation" trajectory. Today's models now reflect the currents in the area, they predict an extreme current in the SouthWest direction. Reachability envelopes in this image are pretty extreme: 20091112_OSSE_Waypoints.jpg. All the white lines are actual plans, except for ru21 (here we decided it was getting too close to the shelf and deleted the first waypoint manually, the updated plan is in 20091112_OSSE_Waypoints_ru21_revision).

• Models Used: UMASSD, NYHOPS, ESPRESSO, COAWST, ENSEMBLE, ENSEMBLE-OBJ produced the plan
• Native Glider Velocity Estimate: 0.25mps
• Surfacing Interval: 3 hours (nocomm trigger)
• Waypoint 'satisfy' radius: 1km

2009-11-11 D-Day

That's D for Demo, at the Baltimore OOI Scientists' meeting. The audience participated in this glider planning session; they decided, with some careful leading on Oscar's part, to "fan out" two of the gliders and unite another two. The original suggestion was to unite UD134 and RU21, but reachability analysis suggested that the only gliders with a reasonable chance of meeting were UD134 and RU05. Thus we decided to join those two, send RU23 down the coast and RU21 northwest (cross-current). This provides RU21 a last chance to try and reach the Hudson Shelf Valley.

Today the new spatially-weighted ensemble model comes on-line, but we still used the original model for the plan.

Storm Warning from David Aragon of an incoming NorEaster: "Right now we are blowing 20-30 knots from the NE, smack into the gliders. This is going to last into Sunday, its going to be very interesting. I am not sure what the plans are for the gliders but each one is going to likely fall at least 20 miles to the SW. Gliders can swim perpendicular to currents and get places but not into them. I think that is what is happening for ru21. I am sending these links because there is a Nor'easter coming and were not going to be able to go anywhere. These are the links we usually use so that all can track."

This is significant because the glider spends a fair bit of time at the surface downlinking data and gets blown around, the models don't account for this extra surface time and so our path predictions may not adequately reflect the NorEastern's influence.

Email correspondence from peggy suggests that the hyperion swath may have been different from that used in planning and that we may not have intersected it after all. It was still a useful test of maneuvering, but we'll have to be careful about interpreting that data.

• Models Used: UMASSD, NYHOPS, ESPRESSO, COAWST, ENSEMBLE-OBJ, ENSEMBLE
  produced the plan
• Native Glider Velocity Estimate: 0.25mps
• Surfacing Interval: 3 hours (nocomm trigger)
• Waypoint 'satisfy' radius: 1km

2009-11-10 Glider Performance, Under the Microscope

Today we continued our path planning experiments with the reachability envelopes. The ENSEMBLE model appeared to be the best overall predictor of glider travel for ud_134 and ru05, which is unsurprising since this was the model we used to plan the glider trajectory and we'd hoped it would be the most accurate anyway. Just because a reachability prediction does not match the glider travel does not mean the forecast was incorrect, because we did not take the specific path dictated by that model.

Many of the predictions were accurate to within the resolution of the planning grid (i.e. a couple of kilometers, after a day's travel) - nice.

There is less variance in model predictions for today. Nevertheless we will continue our testing of reachability predictions. We will send RU21 onward up toward the Hudson valley, RU23 south to 39 18', RU05 and UD_134 will go southwest toward 39 06'. RU23 will go due south to 39 18'.

Today's waypoint lists are manually edited to add an extra waypoint after the runout activity to ensure that the gliders never actually reach the last waypoint in the plan, which should avoid the bug.

From David Aragon:
"Predictions are about 10 days left on battery. We are very science heavy and sampling all the time (ru05 and ru21, the others will last a bit longer). If we want more than 10 days its something to think about (would involve disabling upcast sampling for optics). Glider speed is about 24 km / hr. 10 days is about 250 km if all goes well. Just something to think about in the path/science planning. Well need to be in-shore in that time, the image is what we have as a northern recovery point. The alternative to that is where they were deployed."

Finally, some materials for Baltimore...
Here's a brief video of ru23's dynamic path planner: 20091110_ru23_path.avi
Here's a 3-minute video explaining the path planning process: OSSE_Reachability_Demo.mov
Glider planning accuracy over the experiment thus far: 20091110_Planning_Accuracy_Revised.png

• Models Used: UMASSD, NYHOPS, ESPRESSO, COAWST, ENSEMBLE
  produced the plan
• Native Glider Velocity Estimate: 0.25mps
• Surfacing Interval: 3 hours (nocomm trigger)
• Waypoint 'satisfy' radius: 1km

2009-11-09 Model Comparison

Today the goal is to test the different models' predictions of glider travel against the actual vehicle performance. There are some significant discrepancies in today's forecasts, it appears as though the UMASSD model predicts currents to the North East while most of the other models predict lighter currents or SouthWest currents. The ensemble forecast, averaging in the UMASSD prediction, favors a glider path to the North of ESPRESSO, NYHOPS and COAWST. We will test this by sending gliders along paths of high model variance, and comparing reachability envelopes' accuracy.

The plans are generated with the ensemble model, which shows an interesting feature in the glider operations area at 39.5N, 73.5W.20091109_EnsembleForecast_Feature.png. Peggy established this is from the HOPS UMASSD model, which due to its bathymetry data is the only model that predicts currents in this gridsquare. Thus, it has a disproportionate effect on the glider paths in this region as predicted by the ensemble (basically its the only model operating at that depth).

RU05 is going due South to 39 03' N,
RU23 is headed North East to 73 15' W 39 38' N
RU21 will go explore the Hudson Shelf Valley as a part of a multiple-day excursion (courtesy Pierre). It will cross the Valley 2 or 3 times while staying as far south as possible to keep out of the shipping lanes.
UD_134 will go South West to 73 42' W 39 13' E.

Note that the ru23 plan was uplinked at 7pm Eastern today instead of the scheduled 4pm uplink, as the group took some extra time to make sure the destinations were outside of shipping lanes.

Also, we were notified by David Aragon that some of the glider waypoints didn't take from yesterday, a new bug in the glider where it can get stuck on the last waypoint of a list, even if it's retasked before it reaches the waypoint. This occurs with the waypoint list argument of "0" (take the list in order).

I'm having some trouble producing glider plans with UMASSD in most cases, the currents are too strong and it fails to find a valid path. Overall position predictions from yesterday appear accurate to within a few kilometers.

• Models Used: UMASSD, NYHOPS, ESPRESSO, COAWST, ENSEMBLE produced the plan
• Native Glider Velocity Estimate: 0.25mps
• Surfacing Interval: 3 hours (nocomm trigger)
• Waypoint 'satisfy' radius: 1km

2009-11-08 Come Together

The next goal is to bring the gliders together to one spot on the shelf by Monday. The planner charts a more-or-less direct course. Predictions for yesterday proved pretty good so I don't anticipate needing further changes to velocity estimates.

There was a problem with gliders ru21 and ru05 last night; the waypoints uploaded to the dockserver were ignored by the gliders. Fortunately this didn't significantly disrupt operations, as the old runout waypoints are still more-or-less valid. All the same we're moving up today's uplink to the 1pm surfacing to get the new plans onboard ASAP. Otherwise status is nominal.

For some reason the COAWST model is still failing to find valid paths, could we be out of its domain?

• Models Used: UMASSD, NYHOPS, ESPRESSO, ENSEMBLE produced the plan
• Native Glider Velocity Estimate: 0.25mps
• Surfacing Interval: 3 hours (nocomm trigger)
• Waypoint 'satisfy' radius: 1km

2009-11-07 ASPEN/pCASPER testing notes

We ran some tests with Arjuna, simulating today's ru23 run in MOOS-IvP.

I first validated the plan with ASPEN and saved out a pCASPER format plan, and then ran it using a remote connection to the MOOS-IvP database hosted at MIT. pCASPER connected successfully and commanded Arjuna's new glider behaviors at an 8x time warp 20091107_DynamicSim_screenshot2.png. It received position updates from the MOOS-IvP Nav process and modified execution accordingly, first commanding the initial waypoint in the schedule and later revising this with the prescribed fallback waypoint when the glider fell behind the ru23 plan's schedule.

We used a 50km "Op box" for the MOOS-IVP simulation 20091107_DynamicSim_screenshot3.png; this was large enough to encompass the entire plan out to about 48 hours. The plan included the entire time from the last position update through the telecon planning session and subsequent uplink, out to about two days into the future.

While the simulated glider received the waypoint updates, it appears it did not actually move very far during the simulation - only a small fraction of the intended distance. So the overall report is:

• successful exchange of key messages, software interfaces in place
• unresolved issue with motion of simulated glider

2009-11-07 Cowboy Momentum!

Today's title is a phrase borrowed from Oscar's email. RU21 and RU05 arrived at the Hyperion swath on schedule. With Dave Aragon's suggestions the planning software has started to hit its stride. We had good predictive accuracy overall; at uplink time, the gliders were fairly close to predicted locations in the plan:

• glider RU23: 3km error
• glider RU05: 4-5km error
• glider RU21: <2km error
• glider UD_134: 4-5km error

Also the predicted surge from the NW did hit UD_134, check out the currents it experienced: 20071107_ud_134_zcur_map2.jpg (red arrows in the image). These are the strongest of any yet in the OSSE so far.

Here are today's waypoints: 20071107_OSSE_Waypoints.jpg. RU05 and RU21 follow fastest-possible paths SSW to "reunion point." Models predict they'll arrive about uplink time tomorrow, but I haven't supplied a runout "super waypoint" to occupy them afterword... Instead they'll reign in their horses, kick off their boots and cool their heels at the ranch while the others catch up. For both UD_134 RU23 and the planner prefers to travel North/South first and then go East when we've hit the proper latitude.

So far, it seems that there are a few key indicators that suggest when the glider might not travel like the models predict. These are excessively jaggy paths and large differences between the plans prescribed by different models. By this standard, RU23 and UD_134 might both be in for an interesting day!

• Models Used: UMASSD, NYHOPS, COAWST, ESPRESSO, ENSEMBLE produced the plan
• Native Glider Velocity Estimate: 0.25mps
• Surfacing Interval: 3 hours (nocomm trigger)
• Waypoint 'satisfy' radius: 1km

2009-11-06 Preparing for Hyperion

Here's the lowdown: umd_134 is steaming toward the main group. The models predict that in this area currents pushing South West will make progress difficult. The planner generally suggested an irregular zigzag course; we'll see if this scenario plays out tomorrow.

ru23 will continue inland on its cross-shelf trajectory. Both of these will be collecting good cross-shelf data to track the winter bloom with transects. ru05 will loiter in the Hyperion swath, while ru21 will head south-west to try and join the party in the Hyperion footprint. Models suggest we can probably make it just in time.

We decided on a preliminary "reunion point" at 73 30' W, 39 21'N that we would try to achieve by Sunday. This location is subject to revision tomorrow. With the gliders together, we will have a chance to cross-calibrate sensors and could also test current models by comparing paths chosen by different path planners from similar starting conditions.

Today I also exercised the ASPEN connection. The following image shows the complete mission planning desktop used for today.

20091106_Planning_Desktop.png

With the pipes connected to the Rutgers fleet, the OPENDAP models, and the output to the web portal, it's taking me ~20 minutes to finish the current-optimal, validated plans for all the gliders and commit the results. Most of this time is spent playing with alternative waypoints, looking at reachability predictions, and finally double-checking the resulting glider mission plans by eye (paranoia, perhaps).

Qualitatively positioning predictions are improved over previous days so I'll stick with the new glider native speed estimates.

COAWST model seems to not be finding paths, is it down or are we out of its boundary? The ESPRESSO model may also be having hiccups as it generally fails to find valid paths.

• Models Used: UMASSD, NYHOPS, COAWST, ESPRESSO, ENSEMBLE produced the plan
• Native Glider Velocity Estimate: 0.25mps
• Surfacing Interval: 3 hours (nocomm trigger)
• Waypoint 'satisfy' radius: 1km

2009-11-05 Splitting up the fleet

Having shown that we could perform a "fan out" behavior we established new long-term goals...
With RU 21 we'll try to go north of the hyperion target at 39 32'47.76''N, 73 24'45.00'' W, but not too far North. The Hyperion swath is 7x30km, and we'll reunite the gliders there to sample that area concurrently with the imaging. We'll come toether at a spot on the shelf on mid-weekend, trying to keep the gliders within a couple of km of each other. This will also provide an opportunity for sensor intercalibration.

Today though ru05 and ru23 will continue as before, with newcomer ud_134 on manual until it enters deep water tomorrow.

Technical notes: today we dialed back the glider speed estimates slightly to 0.25mps, this has a couple of immediate impacts including slightly more irregular glider paths (better reflecting sensitivity to currents). Additionally we reduced the number of waypoints, holding the glider to widely-spaced waypoints along each straight-line track and also at "inflection points" in the original plan. These were done via hand-editing after the planner ran but tomorrow we will incorporate this into the actual automatic planning process so that waypoints are only provided at inflection points. The idea is to reduce the 'jitter' in glider paths along straight-line segments and minimize the time spent trying to hit the waypoint exactly.

update from David Aragon: "Winds are high. I think thats the lack of progress we are seeing. Same thing as that one night. Good weather coming though. Gliders are throwing heavy datasets into the air so probably losing ground at the surface too."

• Models Used: ESPRESSO, NYHOPS, UMASSD, COAWST, ENSEMBLE produced the plan
• Native Glider Velocity Estimate: 0.25mps
• Surfacing Interval: 3 hours (nocomm trigger)
• Waypoint 'satisfy' radius: 1km

2009-11-05 (1500 UTC) Crash and Recovery

There was a glider abort last night, at approximately 2009-11-05 04:00 UTC when the ru21 compass failed. David Aragon notified me that the reset would cause the glider to return to the start of its waypoint list after finishing the current waypoint in progress. If we permitted it to happen this would have resulted in considerable backtracking.

This was a good chance to test our on-the-fly replanning. I ran the standard update-and-replan procedure, regenerating the plan for the next uplink opportunity. The replanning moves toward our original goal (NW to begin the cross-shelf line) but taking the new uplink time and position into account. The total replanning turnaround time was approximately 5 minutes. The new plan makes some adjustments to the original trajectory, resulting in a path slightly to the East of the original.

The uplink was scheduled to occur at 16:00 UTC. The glider appeared 5 minutes early, <1.5km from the scheduled uplink location. The new waypoints appear to have been uploaded successfully.

• Models Used: ENSEMBLE produced the plan
• Native Glider Velocity Estimate: 0.3mps
• Surfacing Interval: 3 hours (nocomm trigger)
• Waypoint 'satisfy' radius: 1km

2009-11-04 Backward the Light Brigade

Now making good progress toward the shelf boundary. The uplink location was still a bit ambitious, with gliders not quite at the start of the plan when the trajectory was scheduled to begin. Nevertheless they seem to be tracking trajectories fairly well. Ru05 finished the plan within the resolution of the waypoint cycle, and ru21 as well (granted was running the wrong waypoints but the basic length was appropriate).

There was a discussion with Yi Chao about why the planned trajectories were similar for different models, even when their current predictions differed significantly. My first thoughts are that it is largely due to our short planning timeframe and relatively large grid size (i.e. it's rare that current differences are large and persistent enough to justify a detour on the scale of a 5-km grid square).

There seems to be some evidence of gliders missing waypoints (ru05 and ru21) and backtracking (ru23), I'm discussing with David Aragon whether this warrants a faster surfacing interval to remedy. In the short term, we'll be a bit more conservative with glider speeds and scale back to 0.3m/s. This doesn't appear to effect path locations much but it does affect the predicted times, this will give us a safe way to test the change.

Today's planning session incorporates AUVs as well. The operations area is:

• Datum: 39o 25' 06.82''N, 74o 11' 55.08''W
• Gateway Buoy: 39o 27' 46.17''N, 74o 08' 23.20''W

It was decided that we would sample a biological feature near the shore, moving the original ops box North by one box length if possible. We decided that the gliders would be sent back toward the shore in parallel "cross-shore" lines, spreading them out slightly (Yi) to provide better feedback to the models. For now, this is a bit of a holding pattern; we anticipate one or two of these gliders may continue the cross shelf lines but there are features to the North that may also deserve investigation. A new glider, Delaweae, is incoming and wil be deployed shortly and appear in the North to facilitate this effort.
Today ROMS-ESPRESSO experienced a failure of HF radar, so the data was not available during the daily planning session.

• Models Used: COAWST, NYHOPS, ENSEMBLE produced the plan
• Native Glider Velocity Estimate: 0.3mps
• Surfacing Interval: 3 hours (nocomm trigger)
• Waypoint 'satisfy' radius: 1km

2009-11-03 Take Two

The gliders did not make much forward progress last night (see Yi Chao's report) and we didn't actually make it to the start of the plan at the uplink time. Thus, we'll keep the same goal points and recompute the intermediate trajectories. The preliminary diagnosis is that there were strong subsurface currents which were not reflected in all of the models.

UMASSD is data is not available in the common grid, but the other models are currently running.

Given the glider's current flight patterns I may retune the native glider speed, perhaps to 0.3mps to reflect the slightly slower apparent velocity. Another possible fix would be to incorporate the average time spent on the surface as a bias to increase the relative importance of surface currents in the total "currents felt by glider" calculation for predicting motion.

Models Used: ESPRESSO, COAWST, NYHOPS, ENSEMBLE produced the plan

• Native Glider Velocity Estimate: 0.4mps
• Surfacing Interval: 3 hours (nocomm trigger)
• Waypoint 'satisfy' radius: 1km

addendum on 2009-11-04 We have ascertained that the waypoints sent to ru27 did not take, and that it continued to run its old waypoint list. This highlights the importance of a "superwaypoint" runout action! We'll let it finish today's list since it goes toward the same goal anyway, and then retask on 2009-11-04.

2009-11-02 First Planning Session

It was decided during phone conversations with John Kerfoot, Chip Haldeman and Oscar Schofield that we would make the following changes to operations procedure:

• surface at the 'no-comms' interval of 3-hours
• surface at specific UTC times of 1pm and 4pm Eastern (18:00 and 21:00 UTC) for a position acquisition and uplink, with mission planning happening in the interim.
• use a 'super-waypoint', that is, a distant goal at the end of the waypoint sequence which will serve as a runout action in case the glider finishes the sequence early.

Oscar provided a series of goals to start the first day:

• RU21 39 30' 43.71" 73 10' 13.27"
• RU05 39 19' 10.56" 73 12' 18.05"
• RU23 38 58' 24.59" 73 12' 14.09"

We performed mission planning based on an extrapolated uplink location (i.e. where the glider was probably going to be at 21:00 UTC). The waypoints were emailed to John Kerfoot and uploaded to the glider on time.

Followup: it appears that the gliders began the sequence fairly far from the projected uplink location, and that in fact many of them had drifted away from the intended start position at uplink time. We'll see how much this effects the plan execution. In the future, though, I might not have to be so conservative about placing uplink "start" locations far from the last recorded position.

Oscar suggested that there was a serious weather system coming. Things might get interesting!

• Models Used: ESPRESSO, COAWST, NYHOPS, ENSEMBLE produced the plan
• Native Glider Velocity Estimate: 0.4mps
• Surfacing Interval: 3 hours (nocomm trigger)
• Waypoint 'satisfy' radius: 1km