Algorithms for Distributed and Asynchronous Load Balancing in Multi-Objective Optimization for Robot Autonomy
Abstract
Multi-objective optimization is an e -ective method for decision-making in a robot architecture comprised of distinct, and competing decision-making behavior modules. This work involves advances in the fundamental representation scheme of the interval programming (IvP) model and solution algorithms, with the objective of providing advances in the types of decision making available to unmanned vehicle autonomy systems. In this work new algorithms will be explored to generalize the behavior-based multi-objective optimization model toasynchronously distribute computation load across behaviors. Two classes of techniques will be explored. The ?rst will aim to reduce the calculations for certain kinds of objective functions with plateaus exploitable for e?cient grouping. The second class of algorithms will explore the frequency at which objective functions are created within a particular behavior. These algorithms will exploit cases where the function is found to be predictably similar over iterations, perhaps delaying full function re-generation which is otherwise done on eachiteration of the robot decision-making loop. Overall these methods will reduce the number of low-level maneuver evaluation calculations within a decision-making loop which will be used to o -set an increase in complex vehicle dyamic models. The impact of this work will be improved and faster decision-making on DoD unmanned systems and faster simulation for automated testing of the autonomy system.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Apr 25, 2019
- Source ID
- N000141912180
Entities
People
- Henrik Schmidt
Organizations
- Massachusetts Institute of Technology
- Office of Naval Research
- United States Navy