[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

specimen questions for midterm





Here are a couple of questions:

1. Finite horizon POMDPs are known to be P-Space complete. Given that deterministic STRIPS planning is also P-Space complete, why exactly should we be afraid of POMDP complexity

2. How many conditional plans are considered by POMDP value iteration algorithm at level  2 of a 2-horizon search? (assume there are |a| different actions and |o| distinct percepts).

3. How does the convergence rate of value iteration over an an infinite horizon MDP vary as the discount factor gamma is increased from zero to 1? 

4. In the worst case, how much space do you need to represent the value function of (a) infinite horizon MDP  (b) infinite horizon POMDP