pcooksey / trex2-agent

An architecture for goal directed control of autonomous agents

T-REX:

A deliberative executive for goal directed control

This project implements the t-rex agent that allows the execution of multiple loosely coupled deliberative control loops called reactors.

The general idea is to distribute deliberation (such as automated planning) into simpler control loops called reactors.

In contrast to classical planner+executive found in 3-tiered robotics architectures, the system also tends to truly interleave planning and execution in the sense that often (especially with the europa based deliberative reactor) the execution will interrupt deliberation in order to ensure that the agent is always aware of the world state evolution.

While a reactor is not necessarily based on planning techniques, the timeline based medium of interaction is perfectly fit for timelien based planners and allow the user to abstract tasks into a more high level goal directed control.

Dependencies

The core library of t-rex depends on:

• cmake: http://www.cmake.org
• C++ boost libraries 1.47 or above: http://www.boost.org

And these two are the only strong requirements to install t-rex (although a very bare-bone version of it).

The compilation is also setup by default to compile the europa reactor which in turn requires:

• europa-pso 2.6: https://github.com/nasa/europa

But you can safely disable europa via setting WITH_EUROPA to OFF via cmake tools (ccmake or cmake-gui)

Code documentation (the little there is...) is formated for doxygen: http://www.doxygen.org

Additionally t-rex generates few graphviz files in its log directory. Installing graphviz (http://www.graphviz.org/) or compatible graph visualization tools is a good way to visualize these graphs even if not really required.

Further, optional plug-ins may require extra dependencies; for example the ROS extension if enabled naturally expect ROS to be installed (http://www.ros.org) and so on for others.