Safe Control and Reinforcement Learning in Autonomous Systems

The field of autonomous systems is moving towards developing safer and more reliable control methods, with a focus on integrating reinforcement learning and control theory. Recent research has explored the use of control barrier functions, Hamilton-Jacobi reachability, and linear parameter-varying control to ensure safety and stability in various applications, including aerospace, robotics, and satellite systems. Noteworthy papers in this area include:

  • A Quadratic Programming Approach to Flight Envelope Protection Using Control Barrier Functions, which introduces a new approach to flight envelope protection using control barrier functions.
  • HJRNO: Hamilton-Jacobi Reachability with Neural Operators, which proposes a neural operator-based framework for solving backward reachable tubes efficiently and accurately.
  • Multi-Constraint Safe Reinforcement Learning via Closed-form Solution for Log-Sum-Exp Approximation of Control Barrier Functions, which presents a CBF-based safe RL architecture that effectively mitigates the issues of differentiable optimization and computational complexity.
  • Stability Enhancement in Reinforcement Learning via Adaptive Control Lyapunov Function, which introduces a framework that enhances stability and safety through a task-specific CLF design method, dynamic adjustment of constraints, and improved control input smoothness.

Sources

Unifying Direct and Indirect Learning for Safe Control of Linear Systems

A Quadratic Programming Approach to Flight Envelope Protection Using Control Barrier Functions

Stability Enhancement in Reinforcement Learning via Adaptive Control Lyapunov Function

HJRNO: Hamilton-Jacobi Reachability with Neural Operators

Designing Control Barrier Function via Probabilistic Enumeration for Safe Reinforcement Learning Navigation

Deep Reinforcement Learning Policies for Underactuated Satellite Attitude Control

Multi-Constraint Safe Reinforcement Learning via Closed-form Solution for Log-Sum-Exp Approximation of Control Barrier Functions

Linear Parameter Varying Attitude Control For CubeSats Using Electrospray Thrusters

Built with on top of