Future CABots
- CABot3 is due in May 2009.
- It should have:
- Cognitive Mapping
- Several objects
- Figure ground separation
- Exploration by default
- Labelling
- Two Environments
- PP attachment via use
- We hope it will have:
- Learning plans
- Learning compound visual objects
- Learning PP attachment via use.
- We'll be submitting a follow on grant for the next CABots in
January.
- Our current thinking is that CABot4 will be a conversational
games agent.
- Anyting we didn't finish in CABot3
- Better CAs
- The final CABot of that project (CABot6?) should have
- Learning of the semantics of a reasonably sophisticated
(3-d game) environment (with physics).
- Learning via conversation.
- Learning basic language primitives (e.g. lexical system,
grammar rules)
- Vision with light constancy.
- Physics
- Maybe attention
- We're open to other ideas
- Eventurally we'd like to roll it out to real robots.
