Multi-agent Reinforcement Learning With WarpDrive

This notebook introduces multi-agent reinforcement learning (MARL) with WarpDrive (Lan et al. WarpDrive is a flexible, lightweight, and easy-to-use open-source framework that implements end-to-end deep MARL on GPUs. WarpDrive enables orders-of-magnitude speedups compared to CPU-GPU implementations, using the parallelization capability of GPUs and several design choices to minimize communication overhead. WarpDrive also prioritizes user-friendliness - it has utility functions to easily build MARL environments in CUDA and quality-of-life tools to run end-to-end MARL using just a few lines of code, and is compatible with PyTorch. WarpDrive includes the following resources. code - documentation -, and white paper -

Open in Open In Colab

Give us a ⭐ on Github | Check out the documentation | Join us on Slack


This notebook requires some packages besides pytorch-lightning.

! pip install --quiet "ffmpeg-python" "rl-warp-drive>=1.6.5" "setuptools==59.5.0" "ipython[notebook]" "torch>=1.8" "torch==1.10.*" "torchvision==0.11.*" "torchtext==0.11.*" "torchmetrics>=0.7" "pytorch-lightning>=1.4"

⚠️ PLEASE NOTE: This notebook runs on a GPU runtime. If running on Colab, choose Runtime > Change runtime type from the menu, then select GPU in the ‘Hardware accelerator’ dropdown menu.


This tutorial provides a demonstration of a multi-agent Reinforcement Learning (RL) training loop with WarpDrive. WarpDrive is a flexible, lightweight, and easy-to-use RL framework that implements end-to-end deep multi-agent RL on a GPU (Graphics Processing Unit). Using the extreme parallelization capability of GPUs, it enables orders-of-magnitude faster RL compared to common implementations that blend CPU simulations and GPU models. WarpDrive is extremely efficient as it runs simulations across multiple agents and multiple environment replicas all in parallel and completely eliminates the back-and-forth data copying between the CPU and the GPU during every step. As such, WarpDrive - Can simulate 1000s of agents in each environment and thousands of environments in parallel, harnessing the extreme parallelism capability of GPUs. - Eliminates communication between CPU and GPU, and also within the GPU, as read and write operations occur in-place. - Is fully compatible with Pytorch, a highly flexible and very fast deep learning framework. - Implements parallel action sampling on CUDA C, which is ~3x faster than using Pytorch’s sampling methods. - Allows for large-scale distributed training on multiple GPUs.

Below is an overview of WarpDrive’s layout of computational and data structures on a single GPU. image0 Computations are organized into blocks, with multiple threads in each block. Each block runs a simulation environment and each thread simulates an agent in an environment. Blocks can access the shared GPU memory that stores simulation data and neural network policy models. A DataManager and FunctionManager enable defining multi-agent RL GPU-workflows with Python APIs. For more details, please read out white paper.

The Warpdrive framework comprises several utility functions that help easily implement any (OpenAI-)*gym-style* RL environment, and furthermore, provides quality-of-life tools to train it end-to-end using just a few lines of code. You may familiarize yourself with WarpDrive with the help of these tutorials.

We invite everyone to contribute to WarpDrive, including adding new multi-agent environments, proposing new features and reporting issues on our open source repository.

We have integrated WarpDrive with the Pytorch Lightning framework, which greatly reduces the trainer boilerplate code, and improves training modularity and flexibility. It abstracts away most of the engineering pieces of code, so users can focus on research and building models, and iterate on experiments really fast. Pytorch Lightning also provides support for easily running the model on any hardware, performing distributed training, model checkpointing, performance profiling, logging and visualization.

Below, we demonstrate how to use WarpDrive and PytorchLightning together to train a game of Tag where multiple tagger agents are trying to run after and tag multiple other runner agents. Here’s a sample depiction of the game of Tag with 100 runners and 5 taggers.