Mixed Initiative Machine for Instructed Computing: MIMIC

Period of Performance: 06/06/2007 - 04/06/2008

$70K

Phase 1 STTR

Recipient Firm

Aptima, Inc.
12 Gill Street Array
Woburn, MA 01801
Principal Investigator
Firm POC

Research Institution

Cognitive Engineering Research Institute
5865 South Sossaman Road
Mesa, AZ 85212
Institution POC

Abstract

In less than a decade, UAVs have transitioned from a curiosity on the battlefield to a core capability for intelligence and attack. Recent investment has advanced the state the art in algorithms for routing vehicles and for interpreting the imagery they return. However, the intelligence that drives tactical maneuvering remains firmly lodged in the heads of operators. To capture that knowledge and embed it in control devices requires that these systems learn from human UAV commanders how they generate, select, and execute maneuvers to evade and deceive the enemy. We propose to develop Mixed Initiative Machine for Instructed Computing. MIMIC will learn novel models of tactical decision making and control from well-selected exemplars through interactions with operators. To create MIMIC, we will develop use cases that represent critical decision factors in UAV operations, develop an information representation and model (MIMIC-IR) and a learning management model capable of learning from humans in these use cases, and a prototype user interface for managing machine learning.