Agent
- A lot of that stuff is published elsewhere (see the CABot web site).
- What's not published is the agent itself.
- It's not that we've done vision, language, and learning in isolation.
The same system does them all.
- It also manages planning (using something like a Maes net) and
acting.
- It really does move about in the simulated environment.
- One thing that the agent gives us is nice methodological support.
- In the past we've been criticised for experiments validating
foundational work. "How much of the result is the system, and
how much is the way you present the data?"
- Now we can use the agent to present the data in the way
people receive it.
- The system is also broken into subnets. These can be improved
on, without effecting (much) the rest of the system.
- Other systems, e.g. speech recognition, can be plugged in.
