Why T-REX?
What would anybody want to use T-REX? Here are some of the benefits we hope to deliver with T-REX:
Goal-directed Commanding: Do you want to command a robot with goals telling it what to do, instead of detailing precisely how it should achieve them? If so, then a planning system is useful to turn goals into commands automatically, saving the robot operator alot of work, and reducing the possibility of error.
Robust Execution: Many real-world environments are highly uncertain. If your robot can decide how best to accomplish its objectives based on the situation at hand, then it can be robust to these uncertainties. A planner is a principled way to select suitable actions to achieve a goal, and it can leverage a plethora of work in automated reasoning to do so. Planners running during execution are handy because they can incrementally figure out what to do and/or they can re-plan if prior plans do not work out.
Safe Execution: Many sophisticated robots are expensive, complex machines that should be programmed with care. A model can capture safety constraints that must be observed by all executions. Such a model is directly used by the planner and all plans produced are guaranteed to comply with these constraints. We call this model compliant execution. If the model is correct, this provides a powerful quality assurance mechanism for robot behavior.
High-level Programming: It's easy to get lost in details when programming a complex machine. A high-level modeling language raises the level of abstraction to the essential features of the robot mechanisms, available behaviors, and the task domain. This approach allows a programming model that can focus on the problem domain rather than worry about threads, memory management, and so forth.
Key Ideas
There are a few key ideas behind T-REX, and of course, the devil is in the details. Those details deeply involve EUROPA-2. However, as a first introduction, we present the key concepts.
Constraint-based Temporal Planning
Constraint-based Temporal Planning is a paradigm for representing and reasoning about plans. It is a foundation of T-REX, and is presented here informally.
Timelines-based Execution
Many people are intuitively familiar with the notion of a timeline. In T-REX, a timeline is a core primitive for representing all state in the past, present and future. Check this out if you want to see more formally how this works.
Sense-Plan-Act
Timelines are great, but what we really care about is how we fill them out. A long standing paradigm for robot control sees a control loop as a continuous cycle of Sensing, Planning, and Acting. T-REX applies this simple paradigm, thus placing planning at the core of a control loop in a systematic way. Check this out to learn more.
Divide-and-Conquer
An age old approach for tackling complexity is to divide and conquer. T-REX embraces this idea to enable a planning centric system to scale well, enabling a spectrum of variously reactive and deliberative behavior. Here's how.
T-REX in the Scheme of Things
People have been programming robots for quite some time. If you want to find out more about how approaches to programming robots have evolved, and how T-REX fits in to that work, then this is a good place to go.
