Stochastic Optimization and Learning for Planning in Autonomous Systems
Abstract
Planning in autonomous systems addresses the problem of deciding what to do and when to do it in the management of one, or fleets, of autonomous bots such as drones, unmanned aerial vehicles, mobile sensors, or even cruise missiles. These decisions have to be made over time, in the presence of different sources of uncertainty. Sequential decision problems are the focus of approximately ten major fields of research, along with a number of subfields, each offering their own modeling and algorithmic approaches. This contrasts sharply with the much more unified fields of deterministic optimization, reflecting, we believe, the tremendous diversity of these problems. This diversity exists in the field of autonomous systems, which represents an exceptionally rich class of sequential decision problem in that it combines decisions to collect information which change beliefs, decisions to communicate, as well as decisions that affect the physical system (these are classically kept separate in the academic literature). In addition, these problems introduce rich sources of uncertainty because of the typical lack of knowledge about the environment, in addition to modeling the complex behaviors of adversaries. As a result, we offer a universal modeling framework that covers all of these problems. This framework requires searching over policies (functions for making decisions), where we identify four meta classes of policies that covers every solution approach that has been suggested in the research literature or used in practice. This universal framework has identified a new class of policy, parametric cost function approximations, that bridges stochastic optimization and machine learning which is central to our proposal.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jan 14, 2022
- Source ID
- FA95501910203
Entities
People
- Mengdi Wang
Organizations
- Air Force Office of Scientific Research
- Trustees of Princeton University
- United States Air Force