There are many different microscopic phenomena in the ocean such as blooms and sediment transport processes. Scientists have technical difficulty gaining a mechanistic understanding of these phenomena and tracking their spatial movement. Conventional methods involve ship-based operations of sampling and observing, which requires humans to be present. Now there are more effective methods of exploration, which are conducted by the autonomous underwater vehicles (AUVs). AUVs can host a wide variety of sensors that can decipher the surrounding physical, chemical, and biological environment.
The conventional AUV control systems are composed of reactive approaches that rely on manually scripted plans. They respond to their immediate environment, such as changing the pitch of the vehicle to avoid an incoming obstacle. However, this reactive change may disregard prime opportunities for sampling and observations.
This reactive approach is adequate, but it prevents the AUV from gaining in situ adaptations to take advantage of unexpected opportunities. A safe and effective AUV adaptation requires that there is a balance between mission objectives and maximizing the usage of resources on the vessel. AUV missions will become more complex and lengthy, which means the AUV needs to be more efficient in collecting data while responding to its immediate environment.
MBARI's autonomous team developed an adaptive software system that combines automated planning and probabilistic feature detection. Probabilistic feature detection uses scientific observations to determine the chance of the vehicle finding an object of interest. The automated onboard planning feature allows the AUV to have goal-directed commands within the mission and to replan in the future to take advantage of unexpected opportunities.
The software is called the Teleo-Reactive Executive (T-REX), which is built around the model of artificial intelligence (sense, plan ahead, and act accordingly to find opportunities). T-REX is currently deployed on MBARI’s Dorado-class AUV, which uses its instruments to sense its surroundings, come up with a plan, and then execute the plan.
The architecture of T-REX is able to compute short-reaction times and extended planning time with modules called reactors. The software is able to deliberate mission goals in order to figure out the best method to carry out goals with respect to the surrounding environment and current state of the vehicle (e.g. vehicle may be low on available energy). If the surrounding environment does not match the mission plan, T-REX will automatically form a new plan to compensate for this sudden shift.
Reactors contain a control loop that is dependant on observations of the surrounding environment. The control loops contain information necessary to complete the mission objectives, and the T-REX software simultaneously coordinates all of the control loops.
The Dorado-class AUV has three reactors: a mission manager which plans mission goals, a navigator that determines the vehicle’s navigation, and an executive which controls the low-level functional components of the vehicle. All of the reactors must have the same unit of time (tick) in order to synchronize with each other by a certain time period. Within that time period, the reactors are free to deliberate to determine the optimal mission goal before the next tick.
Reason about system state
The T-REX uses one domain model for all the reactors which details the vehicle’s operational constraints. This model is written in a constructed language that can include units of resources and time periods. Such constraints may include the vehicle having to surface within a time period or the frequency of collecting water samples.
Timelines help to describe the state of the vehicle with intervals of time in which a task can be accomplished. Thus the vehicle can determine the probability of completing a particular task within the time period.
Each reactor contains and controls the evolution of only one timeline. A different reactor can observe an external timeline and suggest new mission goals. This suggestion is then transmitted back to the reactor with the internal timeline which then can make the necessary adjustments. This plan can then impact all of the other timelines owned by the respective reactors.
Within the time periods, if there is a sudden change in a variable, this will impact a planned action. For instance, a sudden change in state may indicate an occurrence such as low battery power or a change in temperature. This unexpected observation will then trigger replanning within the reactor, which will attempt to find a solution. If the reactor does not find a solution, this failure will be passed along the chain of reactors until a new plan is found or the entire system becomes corrupt.
This approach minimizes system degradation, partitions the planning and problem solving process, and allows for software engineering modifications.
Probabilistic feature detection
The ocean phenomena are often unpredictable and difficult to track. There is a need to classify online estimation of ocean features in the water column. Statistical data from previous AUV surveys are used to come up with data that is similar to a significant environmental feature. This data is then integrated with the incoming sensor data in order to find significant environmental features that merit further investigation.
The T-REX has been used for a number of science missions regarding upper-water-column studies. The Dorado-class AUV uses two onboard computers: a main vehicle computer with a “244 megahertz PC/104 stack running QNX Software Systems’ (Ottawa, Canada) QNX real-time operating system” and a 367 megahertz Winsystems Inc. (Arlington, Texas) EPIC EPX-GX500 AMD Geode stack, which operates Red Hat Linux and T-REX.
The ground-breaking application of T-REX has been to find, map, and sample intermediate nepheloid layers (INLs). These are small sheets of suspended sedimentary particles which are brought from the seafloor to the surface.
- The T-REX was able to map and bring back water samples from significant concentrations of the INL occurrences.
- The T-REX has been able to successfully complete a mission objective while sampling in situ features of INL occurrences.
- The T-REX has also been used to detect sediment plumes that had significant concentrations of agricultural runoff flowing into the Monterey Bay. T-REX mapped and used the water sampler to obtain small parts of the plume marked by low salinity and high nitrate signatures.
- In a recent occurrence, T-REX was used along a path of ocean phenomena in the northern Monterey Bay shelf. It was able to analyze the temperature gradients that indicate temperatures in ocean fronts (where the ocean is close to land or buildings).
T-REX is a new open-source robotic controller that allows for in situ sampling.
The long-term goals are increased autonomy of AUVs to navigate increasingly difficult hazards in longer periods of time. There is great potential in having sustained ocean exploration of the water column done by AUVs, which balance mission objectives and usage of resources on the vehicles.
The above information is from the following reference:
Rajan, Kanna, Py, Frederic, and McGann, Connor. "Adaptive Control of AUVs Using Onboard Planning and Execution". Sea Technology and Magazine. 23 Jul, 2010. http://sea-technology.com/features/2010/0410/adaptive_control_auvs.html