This repository provides a comprehensive introductory course on Decision Intelligence using the Proximal Policy Optimization (PPO) algorithm and its extensions. It targets individuals curious about deep reinforcement learning, aiming to equip them with the theoretical understanding and practical coding skills to build decision AI application prototypes efficiently.

How It Works

The course is structured around eight chapters, each focusing on a specific aspect of decision intelligence, from basic concepts to advanced topics like multi-agent systems and sequential modeling. It emphasizes a "one algorithm solves all" philosophy, demonstrating how PPO and its family can address diverse applications. The approach pairs theoretical explanations with corresponding code implementations, facilitating a clear understanding of algorithm logic and practical application.

Quick Start & Requirements

• Installation: No explicit installation instructions are provided for the core course materials. Code examples are likely to be run within Python environments.
• Prerequisites: Python, deep learning frameworks (e.g., PyTorch, TensorFlow, implied by typical DRL implementations), and potentially specific libraries mentioned in chapter content. Access to Bilibili for video lectures is recommended.
• Resources: Course materials include PPTs, manuscripts, QA summaries, homework, solutions, supplementary algorithm theory, and demo code for each chapter. Datasets are available on HuggingFace.

Highlighted Details

• Covers a wide range of DRL topics: complex action spaces, multi-modal observations, sparse rewards, temporal modeling, multi-agent systems, and RLHF.
• Provides a direct mapping between algorithm theory and practical code examples for each concept.
• Offers supplementary materials like algorithm derivations and QA summaries for deeper understanding.
• Includes application examples for various domains such as rocket recovery, soft robotics, and autonomous driving.

Maintenance & Community

The project is actively updated since December 2022. Community interaction is facilitated via a WeChat assistant, Slack, GitHub Issues, and social media channels (Bilibili, Zhihu, YouTube).

Licensing & Compatibility

Released under the Apache 2.0 license, which permits commercial use and integration with closed-source projects.

Limitations & Caveats

The README does not specify hardware requirements (e.g., GPU) for running the provided code examples, which may be necessary for practical DRL training. The course is described as introductory, so advanced users might find the depth limited.