From-scratch implementation of Proximal Policy Optimization (PPO) for continuous control using the Pendulum-v1 environment. The project focuses on policy/value network design, numerical stability, hyperparameter sensitivity, and performance analysis. Implemented in PyTorch without high-level reinforcement learning frameworks.