人工智能英文版课件：03 Solving Problem by Serching.pptVIP

* Searching with Partial Information * Searching with Partial Information We assumed deterministic action This will be almost the same for nondeterministic The agent does not know the outcome of the action in both cases owing to the absence of sensor Murphy’s Law world Suck action sometimes deposits dirt on the ground only if there is no dirt there already Not well functioning appliances Suck in Initial state {1,2,3,4,5,6,7,8} leads to {1,2,3,4,5,6,7,8} again! The world remain unchanged, the problem is unsolvable * Searching with Partial Information Contingency problems If the agent can obtain new information form its sensors after acting The solutions are usually in tree form Branch depends on new information Assume the vacuum cleaning agent has Position sensor Local dirt sensor The percept [L, Dirty] means that the agent is in either {1,3} The agent may formulate action sequence [S, R, S] * Searching with Partial Information Contingency problems The agent may formulate action sequence [S, R, S] By the Murphy’s Law, the last action may fail Arrived state 8 Sucking on a clean floor leads to deposit of dirt Jump to state 6… No fixed action sequence can guarantee for a solution So, we do not use a fixed action sequence [Suck, Right, if [R, Dirty] then Suck] Action is selected to deal with contingency * Searching with Partial Information Sequential solution Sometimes possible if it is a fully observable Murphy’s Law world Dirt may be produced when sucking on a clean floor We just don’t suck on a clean floor No contingency arise If the floor is already clean, no sucking is needed to achieve a goal * Searching with Partial Information Interleaving We do not need to consider every possibility of contingency that might occur Just start to act see which contingency does arise The agent can continue to solve the problem Taking into account the additional information Useful for exploration and game playing * Thank you * Uninformed Search Stra