Studying human spatial navigation processes using POMDPs

Brian J. Stankiewicz, Matthew McCabe, Gordon E. Legge

Research output: Contribution to conferencePaperpeer-review

1 Scopus citations


Humans possess the remarkable ability to navigate through large-scale spaces, such as a building or a city, with remarkable ease and proficiency. The current series of studies uses uses Partially Observable Markov Decision Processes (POMDP) to better understand how humans navigate through large-scale spaces when they have state uncertainty (i.e., lost in a familiar environment.). To investigate this question, we familiarized subjects with a novel, indoor, virtual reality environment. After familiarizing the subject with the environment, we measured subject's efficiency for navigating from an unspecified location within the environment to a specific goal state. The environments were visually sparse and thus produced a great deal of perceptual aliasing (more than one state produced the same observation). We investigated whether human inefficiency was due to: 1) accessing their cognitive map; 2) Updating their belief vector; or 3) An inefficient decision process. The data clearly show that subjects are limited by an inefficient belief vector updating procedure. We discuss the ramifications of these finding on human way-finding behavior in addition to more general issues associated with decision making with uncertainty.

Original languageEnglish (US)
Number of pages6
StatePublished - Dec 1 2004
Event19th National Conference on Artificial Intelligence - San Jose, CA, United States
Duration: Jul 25 2004Jul 26 2004


Other19th National Conference on Artificial Intelligence
CountryUnited States
CitySan Jose, CA

Fingerprint Dive into the research topics of 'Studying human spatial navigation processes using POMDPs'. Together they form a unique fingerprint.

Cite this