DeepMind Control Suite (2018-01-02T00:00:00.000000Z)

TL;DR

The DeepMind Control Suite is a set of continuous control tasks with a standardised structure and interpretable rewards, intended to serve as performance benchmarks for reinforcement learning agents.

Abstract

The DeepMind Control Suite is a set of continuous control tasks with a standardised structure and interpretable rewards, intended to serve as performance benchmarks for reinforcement learning agents. The tasks are written in Python and powered by the MuJoCo physics engine, making them easy to use and modify. We include benchmarks for several learning algorithms. The Control Suite is publicly available at this https URL A video summary of all tasks is available at this http URL .

Authors

T. Lillicrap

23 papers

Tom Erez

3 papers

Yuval Tassa

5 papers

TL;DR

Abstract

Authors

References37 items

Deep Reinforcement Learning that Matters

A Distributional Perspective on Reinforcement Learning

Learning human behaviors from motion capture by adversarial imitation

Reproducibility of Benchmarked Deep Reinforcement Learning Tasks for Continuous Control

Data-efficient Deep Reinforcement Learning for Dexterous Manipulation

Learning to Perform Physics Experiments via Deep Reinforcement Learning

Layer Normalization

OpenAI Gym

Benchmarking Deep Reinforcement Learning for Continuous Control

Asynchronous Methods for Deep Reinforcement Learning

Prioritized Experience Replay

Continuous control with deep reinforcement learning

Simulation tools for model-based robotics: Comparison of Bullet, Havok, MuJoCo, ODE and PhysX

Human-level control through deep reinforcement learning

Trust Region Policy Optimization

Adam: A Method for Stochastic Optimization

MuJoCo: A physics engine for model-based control

Synthesis and stabilization of complex behaviors through online trajectory optimization

The Arcade Learning Environment: An Evaluation Platform for General Agents

Stochastic Complementarity for Local Control of Discontinuous Dynamics

Real-time reinforcement learning by sequential Actor-Critics and experience replay

Reinforcement Learning Using Neural Networks, with Applications to Motor Control

Dynamic Programming and Optimal Control, Two Volume Set

The swing up control problem for the Acrobot

Evolving virtual creatures

Neuronlike adaptive elements that can solve difficult learning control problems

Distributed prioritized experience replay. Under submission, 2017a. Anonymous. Distributional policy gradients. Under submission

Distributional policy gradients

The database was created with funding from NSF

CMU Motion Capture Database

Dynamic Programming and Optimal Control Volume 1 SECOND EDITION

Termination and Discount: Control problems are classiﬁed as ﬁnite-horizon, ﬁrst-exit and inﬁnite-horizon

Function Optimization using Connectionist Reinforcement Learning Algorithms

Action: With the exception of the LQR domain

Dynamics: While the state notionally evolves according to a continuous ordinary diﬀerential equation ˙ s = f c ( s , a ), in

Reward: The range of rewards in the Control Suite, with the exception of the LQR domain

Observation: The function o ( s , a ) describes the observations available to the learning agent. With the exception of point-mass:hard (see below), all tasks

Field of Study

Journal Information

Name

Volume

Venue Information

Name

Type

URL

Alternate Names