AGENTS THAT PLAN
Memory versus Computation
Memory Based Implementation
- The action is selected by designer
- Agent would require large amounts of memory
- Designer would require foresight in predicting appropriate reactions for all possible situations.
Computation Based Implementation
- It will reduce the agent’s memory, requirements and the burden on the designer:
- The designer specify the computation instead of implementing programs for all possible situations.
- If computations which predict consequences of actions can be automatically learned and evolved, then designer is no need to foresee.
State Space Graphs
State Space Graph: A graph representing all of the possible actions and situations, any path in the graph can be taken to represent a goal situation.
A state space graph is most useful structure for keeping track of the effects of several alternative sequences of action.
The nodes in the graph represent possible states of the world and it can be either in iconic form or feature form. Each arc is labeled with operators.
Example: Consider a world containing three toy block P, Q, R and wanted to stack them so that they are in order P, Q, R from top. Suppose, initially they are on floor, then next state has to be predicted by generating all possible states and selecting appropriate state which leads to goal situation.
In the above diagram the nodes are in iconic form. If the goal situation is (PQR) then two effects in above diagram ((PQ)(R)) and ((QR)) leads to goal situation.
A sequence of operators which are labeled along the path to 2 goal is called a plan. The process of searching such a plan is called planning.
Projection is the process of predicting a particular sequence of world states which resulting from a sequence of actions.
The above Fig. 1.3.2 shows complete state space graph for problem of toy blocks. Here at the initial state (at middle) all the blocks are on the floor. The arcs labeled with operators performs an action leads to next possible world state. The actions can be reversible.
