The examples and demos here show how to use the sensori-motor system. The first set of examples show the basic API, while the second set wires them up to a hand-crafted stimulus-response pipeline, to create simple agents.
While going through these demos, please keep in mind the project goal. This is to craft a sensori-motor API that will make it easy for learning systems to explore, perceive and navigate through the "external" world.
The hand-crafted stimulus-response pipelines are meant to show the general usage pattern that a learning system would learn; its a target that the learning system can aim at.
The API is meant to be an affordance for machine learning, and not for human programmers. If you just wanted to work with files or chat on IRC, you need none of this stuff. All this stuff is complicated and convoluted; it is meant for simple & easy machine interaction, rather than application development.
That is to say, all of these examples are much too complicated and difficult. The eventual goal is to automate the hooking up of sensors to motors. That automation layer is completely missing, just right now. So instead, the list of examples below demo a big complex mess of the guts of the system. If you think this is fugnuts complicated, you're right, it is.
The Architecture Overview provides a sketch of how things should work, eventually (if that day ever comes).
Examples that demonstrate how the API works, and the most basic usage.
file-read.scm-- Stream file contents to StreamValuefile-write.scm-- Stream Atoms/Values to a file.xterm-io.scm-- Stream Atoms/Values fomr/to an interactive terminal.irc-api.scm-- Demo of connecting to IRC and interacting.
Examples showing how prototype agents can be built up in Atomese. These attach the API's (from the previous examples) to Atomese processing pipelines to build crude stimulus-response agents.
xterm-bridge.scm-- Copying text between two xtermsirc-echo-bot.scm-- IRC echo bot demo.filesys.scm-- Demo of navigating a filesystem.
The ultimate design goal is to replace the crude hand-crafted stimulus-response pipelines with a system that learns how to use the sensori-motor interfaces offered here, and use those interfaces to navigate the "external" world.
The demos show what a stimulus-response pipeline looks like, so that learning algos can target that general structure.
Additional demos are provided in the Motor Experiments repo. They are there, not here, because they skip over the connector-mating idea that is central to this repo. Instead of using connectors, those demos are just hand-wired.
Collection of coding hints.
hints.scm-- Using ExecutionOutputLink and GroundedSchema