Safe Reinforcement Learning (SRL) algorithms aim to learn a policy that maximizes the reward while satisfying the safety constraints. One of the challenges in SRL is that it is often difficult to balance the two objectives of reward maximization and safety constraint satisfaction. Existing algorithms utilize constraint optimization techniques like penalty-based, barrier penalty-based, and Lagrangian-based dual or primal policy optimizations methods. However, they suffer from training oscillations and approximation errors, which impact the overall learning objectives. This paper proposes the Permeable Penalty Barrier-based Policy Optimization (P2BPO) algorithm that addresses this issue by allowing a small fraction of penalty beyond the penalty barrier, and a parameter is used to control this permeability. In addition, an adaptive penalty parameter is used instead of a constant one, which is initialized with a low value and increased gradually as the agent violates the safety constraints. We have also provided a theoretical proof of the proposed method's performance guarantee bound, which ensures that P2BPO can learn a policy satisfying the safety constraints with high probability while achieving a higher expected reward. Furthermore, we compare P2BPO with other SRL algorithms on various SRL tasks and demonstrate that it achieves better rewards while adhering to the constraints.

Adjusting the remote electrical tilt (RET) of antennas is one of the important actions targeting run-time optimization of key performance indicators (KPIs) related to service quality in wireless self-organizing networks (SONs). Reinforcement learning (RL) is one of the preferred Machine Learning methods for automating the choice of RET for all the antennas managed by a company in a region. The automated system should ensure that the system will operate within a safe region to maintain a minimum defined service quality. The safe region of operation is typically customizable based on the targeted service quality at any point in time. This customizable nature of the safe region necessitates automated learning of adaptive safety shields for steering the RL agent away from unsafe regions. This paper presents an adaptive safety shield framework that is capable of learning such shields during the training phase of the RL agent. Our adaptive safety shield framework has been evaluated in different RET scenarios, and we have shown the benefits of our proposed framework over the Baseline method currently in use and a vanilla RL-based method in terms of both safety and performance metrics.