This repository implements the PETS (Probabilistic Ensembles with Trajectory Sampling) algorithm from a 2018 paper, combining uncertainty-aware deep network dynamics models with sampling-based uncertainty propagation for reinforcement learning. The code trains RL agents on MuJoCo robotics tasks including CartPole, Reacher, Pusher, and HalfCheetah, achieving model-free-level performance with significantly fewer samples than model-free baselines like SAC and PPO.