Date of Award


Document Type


Degree Name

Master of Science in Computer Engineering


Department of Electrical and Computer Engineering

First Advisor

Gilbert L. Peterson, PhD


Current attempts to prolong the life of a robot on a single battery charge focus on lowering the operating frequency of the onboard hardware, or allowing devices to go to sleep during idle states. These techniques have much overhead and do not come built in to the underlying robotic architecture. In this thesis, battery life is greatly extended through development of a behavior-based power management system, including a Markov decision process power planner, thereby allowing future robots increased time to operate and loiter in their required domain. Behavior-based power management examines sensors needed by the currently active behavior set and powers down sensors not required. Additionally, predictive power planning is made possible through modeling the domain as a Markov decision process in the Deliberator. The planner creates a power policy that accounts for current and future power requirements in stochastic domains. This provides the identification of the ability to use lower-power consuming devices at the start of a goal sequence in order to save power for the areas where higher-power consuming sensors might be needed. Power savings are observed through four simulated robots—no power management, lenient power management, strict power management, and predictive power management—in two case studies: 1) Low sensor intensity environment where robots wander randomly while avoiding obstacles and 2) High sensor intensity environment where robots are required to execute a series of tasks. Testing reveals that in a real life scenario involving multiple goals with multiple sensors, the robot’s battery charge can be extended up to 96% longer when using behavior-based power management with predictive power planning over robots that only rely on traditional power management.

AFIT Designator


DTIC Accession Number