Incremental Learning for Robot Sensing and Control

Period of Performance: 01/21/2010 - 07/21/2010

$100K

Phase 1 STTR

Recipient Firm

SET Assoc. Corp.
1005 N. Glebe Rd.Suite 400
Arlington, VA 22201
Firm POC
Principal Investigator

Research Institution

Carnegie Mellon University
5000 Forbes Avenue
Pittsburgh, PA 15213
Institution POC

Abstract

SET Corporation, together with Carnegie Mellon University''s National Robotics Engineering Center (NREC), will develop a system that leverages state-of-the-art sensing, perception, and machine learning to provide trafficability assessments for UGVs for agricultural, security and military applications. It will consist of a set of proprioceptive and exteroceptive sensors that provide rich data about the UGV s environment in conjunction with a learning system that supports a combined experiential and imitative learning regime. We propose a 6 month Phase I effort to 1) develop the underlying algorithms for a combined incremental experiential and imitative learning system, 2) investigate the appropriate sensor modalities, 3) design the general architecture of the integrated system, and 4) evaluate the methods on real data for real-time feasibility and performance over state-of-the-art. We bring to the table an already existing database of data collected from UGVs with many state-of-the-art sensors, ready-made platforms for integrating any additional sensors identified by the sensor study and collecting data, complementary expertise in sensor technology, a software base of cutting-edge perception methods for the competitive analysis, and the machine learning experience and knowledge in the area of online and semi-supervised learning.