domains: Benchmark Planning Problems

Below is a collection of domains for Markovian decision-making models.

(This page has not yet been completed. Check back later this year.)

 
 
sas.png

Drive

Using real-world road data provided by OpenStreet Map (OSM), this domain requires balancing two objective functions: speed and time spent autonomously driving. Lexicographic Partially Observable Markov Decision Processes (LPOMDPs) naturally capture this problem. They place a lexicographic preferences over the optimization of value functions, and allow for slack in their optimization.

For more information, please see our IJCAI 2015 paper:

Wray, Kyle H. and Zilberstein, Shlomo. "Multi-Objective POMDPs with Lexicographic Reward Preferences." Twenty-Fourth International Joint Conference of Artificial Intelligence (IJCAI), Buenos Aires, Argentina, July 2015.

pal_pomdp.png

Proactive Learning

A POMDP model which solves proactive learning -- a generalized realistic active learning scenario with multiple, imprecise, irresponsive, cost-varying oracles. Applications range from scientific experiment selection to medical diagnosis laboratory test selection.

For more information, please see our AAAI 2016 paper:

Wray, Kyle H. and Zilberstein, Shlomo. "A POMDP Formulation of Proactive Learning." Thirtieth Conference on Artificial Intelligence (AAAI), Phoenix, AZ, USA, February 2016.