Learning-based Planning and Control with Persistent Safety for UAS

Abstract

Safe learning-based planning and control for unmanned aerial systems (UASs) are considered in this project. The proposed work unifies the performance and optimality features availed by some of the machine learning (ML) algorithms with the guarantees provided by a class of robust adaptive controllers. The research effort seeks to advance the state-of-the-art by designing a complete framework from planning to control for autonomous operation of UASs. It is constructed through the following thrusts: i) Riemannian energy based robust adaptive control of uncertain nonlinear systems with Bayesian and deep learning for performance improvement; ii) fast planning and collision checks to generate desired trajectories using a planner-agnostic approach. The proposed work addresses vital constraints encumbering learning-based control, which prevents its use for real-time safety-critical systems. With the understanding that learning-based control cannot be expected to operate at a rate sufficient for safety and can produce erroneous outputs, the proposed work removes ML components from the system's operation. Instead, ML outputs are incorporated only when they are verified and whenever available, which is enabled by the safety net provided by the robust adaptive controller. Thus, the proposed framework decouples safety from learning; only performance and optimality are allowed to depend on learning. Finally, each component of the framework is designed to produce certificates for control, planning, and learning, which can be consolidated into a global certificate verifying the system's safety and performance. Appropriate tools from control theory, differential geometry, and ML are leveraged towards establishing a framework ensuring persistent safety and apriori predictable performance guarantees.

Document Details

Document Type
DoD Grant Award
Publication Date
Jan 21, 2022
Source ID
FA95502110411XX0

Entities

People

  • Naira Hovakimyan

Organizations

  • Air Force Office of Scientific Research
  • United States Air Force
  • University of Illinois Urbana–Champaign

Tags

Fields of Study

  • Computer science

Readers

  • Mathematical Modeling and Probability Theory.
  • Neural Network Machine Learning.
  • Robotics and Automation.

Technology Areas

  • AI & ML
  • AI & ML - Autonomous Systems
  • AI & ML - Machine Learning Algorithms
  • AI & ML - Neural Networks
  • Autonomy
  • Autonomy - Autonomous System Control