On September 21st 2007, the Autonomous System group undertook a full day cruise to further validate the work undertaken so far on T-REX, the onboard deliberative control system on MBARI's CTD AUV platform. The role of planning in this field trial was to reactively generate commands to the Vehicle Control Sub-system based on feedback from the vehicle and to proactively generate plans based on science and operator objectives. The agent dispatches commands to the vehicle and synchronizes updates from the vehicle with the expected state of the plan.
Our objective this time was to execute a full science mission. By doing so we wanted to demonstrate that T-REX can scale for long durations missions, to stress it so it could replan more frequently and to observe the impact of critical situations in term of system performances and mission behavior.
The data below summarizes the two most representative runs on this day. While we undertook some safety checks to ensure that modifications to our code base did not have impact on safet, the duration of the science mission (in excess of 3 hours) limited us to prolonged experimentations.
Our results for this day are summarized as follows.
T-REX reaction to stressing situationsThe goal of this mission was to undertake 3 different transects :
- from westNode to northNode with a priority of 3
- from farWestNode to southNode with a priority of 1
- from eastNode to centerNode with a priority of 2
A Science mission
The objective of this mission was to reproduce a run already
using the nominal prescripted layered-control approach, but doing so
using T-REX. The mission selected with the help of John Ryan of MBARI,
was a volume survey off of MBARI's M0 mooring. The goal was to make a
Yo-Yo survey from the initial point to the pick-up point and then wait
for the pick-up. As our path network had only one path between these
two points, T-REX was able to find this path and then plan and
execute the mission accordingly with insertion of check-in windows
every 1700 to 1800 sseconds consistent with its domain model.
The mission duration was approximately 3 hours and the performance of the system was comparable to what we had seen in prior trials as illustrated in the CPU performance graph above further validating the scalability of the system in terms of plan size and mission duration.
The CPU data indicates 2 spikes in the cost of synchronization. Analysis of the log files indicates that these spikes correspond to plan failures of the form encoutered in the prior sea trial due to squeezing of uncontrollable durations. This time, the agent simply repaired the plan which explains the increased cost of synchronization. This repair capability has been the key to improving the robustness of the system over the prior sea trial.
On the 2D path, one can see the vehicle has interrupted its science survey to localize at the surface on 6 occassions. This is indicated by the blue colored sections which correspond to GPS hits. The mission progresses from the north-west and finishes at the south east. The northward deviation on ascent to localize is a navigational error due to a bug in the model. Not withstanding this error, the vehicle navigation was correct and confirmed that the compass can be relied on to make heading changes without surfacing.
Also shown is the data processed from this run using MBARI's SSDS Data Portal. Many thanks to Mike McCann for handling this.
This mission set a new at-sea endurance record for T-REX by an order of magnitude, with no adverse effects on system performance. Our focus for future missions will emphasize greater application level functionality in pursuit of interesting science observations