Advanced Deep Reinforcement Learning (DRL) explores the theory and practice of training autonomous agents to make intelligent decisions in complex, dynamic, uncertain, and adversarial environments. The course covers core concepts and modern high-performance methods for robotics and autonomous systems, linking techniques to real-world applications such as autonomous aerial combat research, self-driving vehicles, autonomous drones, and robotic platforms ranging from industrial manufacturing cells to state-of-the-art legged and bipedal humanoids (e.g., Boston Dynamics’ Atlas).The course builds a strong foundation in the principles and state-of-the-art methods that power modern DRL. Topics include neural architectures (MLPs, CNNs, RNNs), advanced model-free algorithms (PPO, SAC), model-based DRL (Dreamer, recurrent state-space models, and learned latent dynamics), and attention/Transformer mechanisms for high-dimensional observations and long-horizon credit assignment.Students gain hands-on experience designing single-agent and multi-agent systems for cooperative and competitive tasks, with emphasis on temporal abstraction and latent embeddings to handle uncertainty and partial observability. The course also addresses scalable DRL engineering: standardized environments, experience replay, distributed actor–learner training, and continuous evaluation. Labs and projects use PyTorch, OpenAI Gym and DeepMind Control Suite, and Ray RLlib. Graduates leave prepared to build robust, high-performance DRL agents for research or production.
Course Prerequisite(s)
***Mechanical Engineering students only: Must complete core course first (EN.535.641 Mathematical Methods for Engineers).