Search |  Contact |  SRI Home Do not follow this link, or your host will be blocked from this site. This is a spider trap. Do not follow this link, or your host will be blocked from this site. This is a spider trap. Do not follow this link, or your host will be blocked from this site. This is a spider trap.A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A ASRI International.  333 Ravenswood Avenue.  Menlo Park, CA 94025-3493. SRI International is a nonprofit corporation.

AIC Seminar Series

Reinforcement Learning Approach to Stochastic Planning Problems

Kee-Eung KimInteraction Lab, Samsung Advanced Institute of Technology

Date:  2006-01-30 at 10:00

Location:  EJ228  (Directions)

   Abstract

This talk is concerned with adapting reinforcement learning algorithms for efficiently solving large and complex stochastic planning problems. A planning problem is specified in terms of a model of the environment in which an agent must execute actions to achieve a specified goal or attain a given level of performance. Executing an action results in a transition from one state of the environment to another state and, generally, the agent has to perform a sequence of actions and visit a set of states in order to provide a satisfactory solution to the planning problem. Stochastic planning problems are the case in which the state transitions are specified as probability distributions. Examples of planning problems include path planning in robotics, resource management in logistics, and controlling interface agents in HCI. The class of stochastic planning problems is known to be computationally difficult and consequently the techniques that take advantage of regularities in particular planning domains are necessary. Reinforcement learning algorithms have been applied to various models of planning problems, and their limitations and issues have been discovered. Specifically, when the agent is confronted with noisiness or enormous amount of observations, traditional approach of naively applying reinforcement learning techniques is likely to fail. This phenomenon is critical in the sense that noisy and abundant observations are inherent in real-world environments – e.g., robots with a lot of sensors that sometimes fail to return accurate values. I will present a new framework for adapting reinforcement learning algorithms to overcome these issues, and show the details of adaptation for specific planning problem models. I will also demonstrate the effectiveness of these algorithms on a number of interesting planning problems by showing experimental results.

   Note for Visitors to SRI

Please arrive at least 10 minutes early as you will need to sign in by following instructions by the lobby phone at Building E. SRI is located at 333 Ravenswood Avenue in Menlo Park. Visitors may park in the parking lots off Fourth Street. Detailed directions to SRI, as well as maps, are available from the Visiting AIC web page. There are two entrances to SRI International located on Ravenswood Ave. Please check the Builing E entrance signage.

SRI International
©2014 SRI International 333 Ravenswood Avenue, Menlo Park, CA 94025-3493
SRI International is an independent, nonprofit corporation. Privacy policy