Towards Deliberative Control in Marine Robotics

We describe a general purpose artificial-intelligence-based control architecture that incorporates in situ decision making for autonomous underwater vehicles (AUVs). The Teleo-reactive executive (T-REX) framework deliberates about future states, plans for actions, and executes generated activities while monitoring plans for anomalous conditions. Plans are no longer scripted a priori but synthesized onboard with high-level directives instead of low-level commands. Further, the architecture uses multiple control loops for a “divide-and-conquer” problem-solving strategy allowing for incremental computational model building, robust and focused failure recovery, ease of software development, and ability to use legacy or nonnative computational paradigms. Vehicle adaptation and sampling occurs in situ with additional modules which can be selectively used depending on the application in focus. Abstraction in problem solving allows different applications to be programmed relatively easily, with little to no changes to the core search engine, thereby making software engineering sustainable. The representational ability to deal with time and resources coupled with Machine Learning techniques for event detection allows balancing shorter term benefits with longer term needs, an important need as AUV hardware becomes more robust allowing persistent ocean sampling and observation. T-REX is in regular operational use at MBARI, providing scientists a new tool to sample and observe the dynamic coastal ocean.