Risk-Sensitive Planning

Methods for planning in stochastic domains often aim for finding plans that minimize expected execution cost or maximize the probability of goal achievement. This implies a neutral attitude towards risk. People, however, are usually not risk-neutral. A gambler, for example, is willing to accept a plan with a smaller expected reward if the uncertainty is increased. Consequently, I have developed a mechanism to incorporate risk-sensitive attitudes into existing probabilistic AI planners, thus extending their functionality.

Relevant publications:

Check out my home page and my other projects!

Sven Koenig


skoenig+@cs.cmu.edu / Last update: January 1 1996