Training reinforcement learning (RL) policies for bipedal robots is a complex and resource-intensive task that requires careful design of reward functions, effective sim-to-real transfer, and rigorous evaluation across diverse environments. Traditionally, these challenges have been addressed with significant human oversight. To alleviate this burden, we propose BiPiL, a novel framework designed to automate the training and deployment of bipedal robots. This structure distinctively combines reinforcement learning, adaptive PID control, and imitation learning to improve the training procedure. By incorporating imitation learning, BiPiL leverages expert demonstrations to accelerate policy learning, improving both sample efficiency and the ability to generalize across different tasks. Furthermore, the system persistently oversees and assesses performance to confirm that the acquired policies remain secure and efficient in practical applications. We demonstrate the effectiveness of BiPiL through extensive tests on both simulated and real robots, showing how it reduces the need for human intervention while maintaining high levels of performance and safety.