Navigating like a Human: Generating Landmark Designations for an Autonomous System using a Novelty Algorithm
Abstract
To improve robotic navigation, and drawing on inspiration from human navigation, a landmark designation algorithm was developed and tested using a simulated environment and in the real world. The algorithm used laser data or visual data to determine whether a new environment was considered novel (i.e., different from the previous data), then used this designation for a landmark. The landmark algorithm was tested to determine 1) if it consistently designated the same locations as landmarks and 2) if landmark designation was sparse enough for practical use. Results showed that the algorithm generated relatively consistent landmarks that also corresponded to human landmark designations, but it could be improved by adding top-down contextual information to be more useful.
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 01, 2018
- Accession Number
- AD1063885
Entities
People
- Sean M. Mcghee
- Troy D. Kelley
Organizations
- United States Army Research Laboratory