Responsive Safety in Reinforcement Learning by PID Lagrangian Methods
This addresses the issue of safety violations in reinforcement learning training for applications requiring reliable constraint adherence, though it is an incremental improvement over existing Lagrangian methods.
The paper tackled the problem of oscillations and overshoot in Lagrangian methods for safe reinforcement learning, which cause constraint violations during training, by introducing a PID Lagrangian method that uses derivatives of the constraint function to improve learning dynamics. The result was a new state-of-the-art performance on the Safety Gym benchmark, with improved hyperparameter robustness and nearly simple implementation.
Lagrangian methods are widely used algorithms for constrained optimization problems, but their learning dynamics exhibit oscillations and overshoot which, when applied to safe reinforcement learning, leads to constraint-violating behavior during agent training. We address this shortcoming by proposing a novel Lagrange multiplier update method that utilizes derivatives of the constraint function. We take a controls perspective, wherein the traditional Lagrange multiplier update behaves as \emph{integral} control; our terms introduce \emph{proportional} and \emph{derivative} control, achieving favorable learning dynamics through damping and predictive measures. We apply our PID Lagrangian methods in deep RL, setting a new state of the art in Safety Gym, a safe RL benchmark. Lastly, we introduce a new method to ease controller tuning by providing invariance to the relative numerical scales of reward and cost. Our extensive experiments demonstrate improved performance and hyperparameter robustness, while our algorithms remain nearly as simple to derive and implement as the traditional Lagrangian approach.