AI-Optimizer

AI-Optimizer is a next-generation deep reinforcement learning suit, providing rich algorithm libraries ranging from model-free to model-based RL algorithms, from single-agent to multi-agent algorithms. Moreover, AI-Optimizer contains a flexible and easy-to-use distributed training framework for efficient policy training.

AI-Optimizer now provides the following built-in libraries, and more libraries and implementations are coming soon.

• Multiagent Reinforcement learning
• Self-supervised Representation Reinforcement Learning
• Offline Reinforcement Learning
• Transfer and Multi-task Reinforcement Learning
• Model-based Reinforcement Learning

Multiagent Reinforcement Learning (MARL)

The Multiagent RL repo contains the released codes of representative research works of TJU-RL-Lab on Multiagent Reinforcement Learning (MARL). The research topics are classified according to the critical challenges of MARL, e.g., the curse of dimensionality (scalability) issue, non-stationarity, multiagent credit assignment, exploration-exploitation tradeoff, and hybrid action. To solve these challenges, we propose a series of algorithms from a different point of view. A big picture is shown below.

Offline-rl-algorithms (Offrl)

Current deep RL methods still typically rely on active data collection to succeed, hindering their application in the real world especially when the data collection is dangerous or expensive. Offline RL (also known as batch RL) is a data-driven RL paradigm concerned with learning exclusively from static datasets of previously-collected experiences. In this setting, a behavior policy interacts with the environment to collect a set of experiences, which can later be used to learn a policy without further interaction. This paradigm can be extremely valuable in settings where online interaction is impractical.

This repository contains the codes of representative benchmarks and algorithms on the topic of Offline Reinforcement Learning. The repository is developed based on d3rlpy(https://github.com/takuseno/d3rlpy) following MIT license, while inheriting its advantages, the additional features include (or will be included):

• A unified algorithm framework with more algorithms
  • REDQ
  • UWAC
  • BRED
  • …
• More datasets:
  • Real-world industrial datasets
  • Multimodal datasets
  • Augmented datasets (and corresponding methods)
  • Datasets obtained using representation learning (and corresponding methods)
• More log systems
  • Wandb

Self-supervised Reinforcement Learning (SSRL)

SSRL repo contains the released codes of representative research works of TJU-RL-Lab on Self-supervised Representation Learning for RL.

❓ Problem to Solve

Since the RL agent always receives, processes, and delivers all kinds of data in the learning process (i.e., the typical Agent-Environment Interface), how to properly represent such "data" is naturally the key point to the effectiveness and efficiency of RL.

In this branch, we focus on three key questions as follows:

• What should a good representation for RL be? (Theory)
• How can we obtain or realize such good representations? (Methodology)
• How can we making use of good representations to improve RL? (Downstream Learning Tasks & Application)

⭐️ Core Idea

Taking Self-supervised Learning (SSL) as our major paradigm for representation learning, we carry out our studies from four perspectives: State Representation, Action Representation, Policy Representation, Environment (and Task) Representation.

These four pespectives are major elements involved in general Agent-Environment Interface of RL. They play the roles of input, optimization target and etc. in the process of RL. The representation of these elements make a great impact on the sample efficiency, convergence optimality and cross-enviornment generalization.

💦 Contribution

The central contribution of this repo is A Unified Algorithmic Framework (Implementation Design) of SSRL Algorithm. The framework provides a unified interpretation for almost all currently existing SSRL algorithms. Moreover, the framework can also serve as a paradigm when we are going to devise new methods.

Our ultimate goal is to promote the establishment of the ecology of SSRL, which is illustrated below.

Towards addressing the key problems of RL, we study SSRL with four types of representations. For researches from all four pespectives, a unified framework of algorithm and imeplementation serves as the underpinnings. The representations studied from different pespectives further boost various downstream RL tasks. Finally, this promotes the deployment and landing of RL in real-world applications.

See more here.

🌏 Intention & Impact

With this repo and our research works, we want to draw the attention of RL community to studies on Self-supervised Representation Learning for RL.

For people who are insterested in RL, our introduction in this repo and our blogs can be a preliminary tutorial. For cutting-edge RL researchers, we expect that our research thoughts and proposed SSRL framework can open up some new angles for future works on more advanced RL. For RL participators (especially who work on related fields), we hope that our opensource works and codes may be helpful and inspiring.

We are also looking forward to feedback in any form to promote more in-depth researches.

Transfer and Multi-task Reinforcement Learning

Recently, Deep Reinforcement Learning (DRL) has achieved a lot of success in human-level control problems, such as video games, robot control, autonomous vehicles, smart grids and so on. However, DRL is still faced with the sample-inefficiency problem especially when the state-action space becomes large, which makes it difficult to learn from scratch. This means the agent has to use a large number of samples to learn a good policy. Furthermore, the sample-inefficiency problem is much more severe in Multiagent Reinforcement Learning (MARL) due to the exponential increase of the state-action space.

Solutions

• Transfer RL which leverages prior knowledge from previously related tasks to accelerate the learning process of RL, has become one popular research direction to significantly improve sample efficiency of DRL.

• Multi-task RL, in which one network learns policies for multiple tasks, has emerged as another promising direction with fast inference and good performance.

This repository contains the released codes of representative benchmarks and algorithms of TJU-RL-Lab on the topic of Transfer and Multi-task Reinforcement Learning, including the single-agent domain and multi-agent domain, addressing the sample-inefficiency problem in different ways.

See more here.

Model-based Reinforcement Learning (MBRL)

This repo contains a unified opensource code implementation for the Model-Based Reinforcement Learning methods. MBRL-Lib provides implementations of popular MBRL algorithms as examples of using this library. The current classifications of the mainstream algorithms in the modern Model-Based RL area are orthogonal, which means some algorithms can be grouped into different categories according to different perspectives. From the mainstream viewpoint, we can simply divide Model-Based RL into two categories: How to Learn a Model and How to Utilize a Model.

• How to Learn a Model mainly focuses on how to build the environment model.

• How to Utilize a Model cares about how to utilize the learned model.

Ignoring the differences in specific methods, the purpose of MBRL algorithms can be more finely divided into four directions as follows: Reduce Model Error、Faster Planning、 Higher Tolerance to Model Error 、Scalability to Harder Problems. For the problem of How to Learn a Model, we can study on reducing model error to learn a more accurate world model or learning a world model with higher tolerance to model error. For the problem of How to Utilize a Model, we can study on faster planning with a learned model or the scalability of the learned model to harder problems.

We have collected some of the mainstream MBRL algorithms and made some code-level optimizations. Bringing these algorithms together in a unified framework can save the researchers time in finding comparative baselines without the need to search around for implementations. Currently, we have implemented Dreamer, MBPO,BMPO, MuZero, PlaNet, SampledMuZero, CaDM and we plan to keep increasing this list in the future. We will constantly update this repo to include new research made by TJU-DRL-Lab to ensure sufficient coverage and reliability. What' more, We want to cover as many interesting new directions as possible, and then divide it into the topic we listed above, to give you some inspiration and ideas for your RESEARCH. See more here.

Contributing

AI-Optimizer is still under development. More algorithms and features are going to be added and we always welcome contributions to help make AI-Optimizer better. Feel free to contribute.