Version 1 and Beyond
- We hope to have a working fLIF agent by March. That's
version 0.5.
- We hope to have an agent with more functionality by the
end of year 1 (July 31st 2007).
- We hope to add locations to the system between 0.5 and 1.
- This should enable the system to handle more than one object.
- It involves the type/token distinction.
- It will be a sort of cognitive mapping.
- This should also enable the agent to remember where things are.
- Note that the V1 system will do very little learning. It
will only do variable binding, and ground words in concrete
CAs.
- We then hope to have agents ready on July 31st 2008 (v2), and
May 30th 2009(v3).
- We're not entirely certain what we want to do with V2.
- We will use a more sophisticated game. Probably one of the
DeathMatch variety.
- I'm really interested in implementing a good parser.
- However, for the rest of the group, I think learning will be
really important.
- We should be able to learn new objects and new words.
- We might be able to learn new plans, and predict opponent
plans.
- Other obvious specialties could be motion, tracking and
aiming.
- I also expect there will be a lot of active vision work.
- We're even less certain about V3.
- We do want to have it wrapped up by the end of May to give us all
June and July to run experiments and write up.
