Learning Autonomy in Synthetic Environments (LASE)

Abstract

The Learning Autonomy in Synthetic Environments (LASE) program will develop simulation-to-simulation and simulation-to-real neuro-symbolic transfer learning techniques that enable more fully unmanned operations by autonomous systems. The autonomy levels of unmanned systems of today are limited because it is assumed that the modeling and simulation (M&S) training environment captures all the relevant phenomena at a high fidelity, when in reality it does not account for the data domain shift common when translating simulation outcomes from the M&S environment to the real world. The LASE approach will integrate symbolic structures (to capture discrete symbolic phenomena like mission objectives) and neural structures (to generalize and encode high-dimensional phenomena like sensor signals, imagery, etc.) to more realistically transfer learned autonomy from a M&S environment. Furthermore, since it is often difficult to choose an appropriate M&S training environment, LASE will also explore the development of a neuro-symbolic digital twin for use in training. LASE transfer of M&S-based learning will enable higher levels of autonomy for systems that operate in environments where command, control, and communications can be restricted or denied.

Document Details

Document Type
Accomplishment
Publication Date
Oct 01, 2024
Source ID
b2fc5ac2640dbcf94419a9b12aa2391f

Tags

Readers

  • Artificial Intelligence
  • Distributed Systems and Data Platform Development
  • Unmanned Aerial System (UAS) Autonomous Capabilities and Mission Reconnaissance.

Technology Areas

  • Autonomy
  • Autonomy - Human-Robot Interaction

Related Documents