SURREAL

Distributed Reinforcement Learning and Robot Manipulation Benchmark

GitHub Paper

Distributed Reinforcement Learning

Open-source, reproducible, scalable distributed reinforcement learning framework



Our goal is to make Deep Reinforcement Learning accessible to everyone. We introduce Surreal, an open-source, reproducible, and scalable distributed reinforcement learning framework. Surreal provides a high-level abstraction for building distributed reinforcement learning algorithms. We implement our distributed variants of PPO and DDPG in the current release. Click to see detailed documentation!

Surreal Repo

Surreal Robotics Suite

Standardized and accessible robot manipulation benchmark with the MuJoCo physical engine



Modular

Programmatically creating new objects and new tasks

Multimodal

Multimodal observations from states to pixels

Control Modes

Joint velocity and end-effector controllers

Teleoperation

Teleoperation utility for demonstration collection

RoboSuite Repo

Surreal Full Stack

Four-layer computing infrastructure for distributed learning



Surreal

Algorithms

Implements distributed reinforcement learning algorithms

Caraml

Protocol

Implements robust communication directives for distributed computing

Symphony

Orchestrator

Uses cloud APIs to allocate resources and replicate the networking topology

Cloudwise

Provisioner

Guarantees reproducibility of cluster setup across cloud providers (GCloud, Azure, etc.)

Meet the Surrealists!

Publication



SURREAL: Open-Source Reinforcement Learning Framework and Robot Manipulation Benchmark.
Linxi Fan*, Yuke Zhu*, Jiren Zhu, Zihua Liu, Anchit Gupta, Joan Creus-Costa, Silvio Savarese, Li Fei-Fei
Conference on Robot Learning 2018

Paper Copy Bibtex