f10-belief-space

# f10-belief-space - Partial Observability(State Uncertainty...

This preview shows pages 1–12. Sign up to view the full content.

Partial Observability (State Uncertainty) Assume non-determinism  Atomic model (for belief states and sensing actions)  Factored model (Progression/Regression) Allow distributional information POMDPs

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Beyond Classical Search § Non-Deterministic Actions § Transition model – Result(s,a) is no longer a singleton § Plans have to be “contingent” § Suck; if state =5 then [Right, Suck] else [] § Why “And nodes”? § Non-cyclic vs. Cyclic § Partial Observability § Is planning actually possible with no observation? § Manufacturing; Compliant motion § Belief-Space search § State repetition § Difficulty is the size of the belief states § Factoring to rescue? § http://rakaposhi.eas.asu.edu/dan-jair-pond.pdf § (Next reading) § Observations § States give out “percepts” that can be observed by actions § Observations partition the belief state § State estimation How does this all connect to MDPs?
Partial Observability § The agent doesn’t quite know its current state § Orthogonal to the action uncertainty § Search is in the space of “sets” of states § If you have no distributional information, then there are 2s states § If you have distributional information, then there are _____ states [POMDPs] § How does the state uncertainty get resolved? § By actions § By (partial) observations § Observations § States give out “percepts” that can be observed § Observations partition the belief § The agent now has a slew of new State Estimation problems § Using sensing and action outcomes to figure out § what state it currently is in “state estimation”/ “filtering” § what state it will get to if doesn’t do anything further “prediction” § what state did it start from based on its knowledge of the current state “smoothing” § ..And planning problems § Plan without any sensing § “Conformant” planning § Plan with a commitment to use sensing during execution § “Contingency Planning” § Interleaved sensing and We did a whole lot of discussion around this single slide; see the lecture video. .

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Always executable actions How does the Cardinality of belief State change? Why not stop as soon as goal state is in the belief state?
“Conformant” Belief-State Search

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Heuristics for Belief Space Search?
Not every state may give a percept; will have to go to a neighbor that does. .

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Using Sensing During Search
State Estimation…

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Generality of Belief State Rep Size of belief states during Search is never greater than |BI| Size of belief states during search can be greater or less than |BI|
State Uncertainty and Actions § The size of a belief state B is the number of states in it. § For a world with k fluents, the size of a belief state can be between 1 (no uncertainty) and 2k (complete uncertainty).

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
This is the end of the preview. Sign up to access the rest of the document.

{[ snackBarMessage ]}

### Page1 / 75

f10-belief-space - Partial Observability(State Uncertainty...

This preview shows document pages 1 - 12. Sign up to view the full document.

View Full Document
Ask a homework question - tutors are online