|
|
RAICA - Abstract
This project is aimed to design and implement a completely new kind of self developing artificial intelligence for robot control.
Robot controlling AIs for non trivial tasks typically use a two-way approach nowadays:
Step 1: Learn from others: Getting teached or pick up movements
Step 2: Imitate: Learn from own experience with the real world
These two steps can follow each other repeatedly.
Today, the knowledge available from own experiences is rarely used. Typically it is used
to optimize some parameters of the associated action or the current belief state, and all
other knowledge included in that experience is abandoned.
In RAICA the AI stores all its experiences in a way, that it can later remember each and every one in every detail.
Based on these recalls, it can improve its picture of the world, by validating new theories of causality
against these memories.
RAICA is being implemented in JAVA. The publication as open source is planed, but takes
at lot longer than expected, maybe by the end of 2014. The official project homepage is
www.roboshock.de
|
|
|