NLBAC: A neural ODE-based algorithm for state-wise stable and safe reinforcement learning

Liqun Zhao, Keyan Miao, Hongpeng Cao, Konstantinos Gatsis, Antonis Papachristodoulou

Abstract: Highlights•An augmented Lagrangian-based method for stable and safe RL with efficient training.•An extension with learned barrier certificates, supported by theoretical guarantees.•Investigation into the infeasibility due to the presence of multiple constraints.•An algorithm using neural ODEs for improved modeling performance with public code.