On 9/7/06, Fredrik Heintz <[EMAIL PROTECTED]> wrote:
> I haven't studied G0 in detail, but one of our current research
> problems is the execution and monitoring of plans. We have one of the
> worlds fastest and most expressive planners, TALplanner, which is
> forward-chaining domain-dependent planner using control rules written
> in a temporal action logic which is very expressive. The problem is to
> take these plans and actually execute them. Since the planning domain
> is by necessity an abstraction compared to the real world there will
> always be potential failures waiting around the corner. For example
> actions may fail in unexpected, or even expected ways; actions may
> take longer or shorter time than expected which affects the execution
> of the rest of the plan; and new high value oppertunities may arise
> that the agent ought to take advantage of. We have focused mainly on
> the first and second item. Our current idea is to annotate the plan
> operators with temporal logical formlas which express safety and
> progress conditions, i.e. conditions which are never allowed to be
> violated and conditions which are only violated if no progress is
> made. Then these formulas are evaluated as the world develops using a
> technique called progression, which is basically partial evaluation of
> the formulas over one state at the time. If a formula is violated we
> currently only replan from the new situation. The work is not
> finished, but we expect to have a paper ready by the end of the month.
> problems is the execution and monitoring of plans. We have one of the
> worlds fastest and most expressive planners, TALplanner, which is
> forward-chaining domain-dependent planner using control rules written
> in a temporal action logic which is very expressive. The problem is to
> take these plans and actually execute them. Since the planning domain
> is by necessity an abstraction compared to the real world there will
> always be potential failures waiting around the corner. For example
> actions may fail in unexpected, or even expected ways; actions may
> take longer or shorter time than expected which affects the execution
> of the rest of the plan; and new high value oppertunities may arise
> that the agent ought to take advantage of. We have focused mainly on
> the first and second item. Our current idea is to annotate the plan
> operators with temporal logical formlas which express safety and
> progress conditions, i.e. conditions which are never allowed to be
> violated and conditions which are only violated if no progress is
> made. Then these formulas are evaluated as the world develops using a
> technique called progression, which is basically partial evaluation of
> the formulas over one state at the time. If a formula is violated we
> currently only replan from the new situation. The work is not
> finished, but we expect to have a paper ready by the end of the month.
OK =) That sounds pretty sophisticated. As you know I'm just beginning to learn about acting/planning, my early focus is on "the other side" of cognition, namely as a truth-maintenance system; but obviously an AGI cannot be complete without acting/planning. So I will work on integrating that into the framework as well.
Give me some time...
YKY
To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
