montrealrobotics / DeepRLInTheWorld

From search engines, to science, to robotics, this reposity is meant to showcase the use of reinforcement learning in the world..

Deep Reinforcement Learning in The Real World

This repository lists the many places deep reinforcement learning is used in the world. Feel free to contribute additions to this list, and we will filter through them and accept the ones that seem to fit the criteria of real-world reinforcement learning.

The list

For each item, please include two parts, a link to an article or video to make the application clear and easy to read for the average person. Second, include a link to the paper or technical report (preferably free) that provides further details and its relation to other work (if it is available). See the posts below for examples of this format.

• Deep reinforcement learning to personalize gamer experience better at Zenga (video game company).

  • Link to presentation

• How Telus (telecomunications company) Applies Model-Based Reinforcement Learning to Reduce Energy Consumption in Data Rooms.

  • Link to presentation

• Deepmind and YouTube use Deep Reinforcement Learning to optimize video compression.

  • Link to Blog, Link to arXiv

• Reinforcement Learning to reduce the operating cost of Google data centres by 40%.

  • Link to Blog
  • Updated work allows for use on more types of datacenters with better performance.

• Kindred AI (AI-robotics company) uses Deep Reinforcement Learning to build robots for grasping and manipulation (SORT)

  • Link to Blog, Link to Trailer, Link to Product: 1, 2
  • Link to presentation

• Training stratospheric balloons using deep RL

  • Link to nature article

• Controlling magnetic fields in nuclear fusion reactors

  • Link to nature article

• Using Reinforcement learning for better text summarization

  • link to project

• Using deep reinforcement learning to help balance gameplay for players at Riot Games

  • link to talk

• Siemens has been using RL in specific controlled use cases for many years.

  • link to slide deck

• Foobot uses deep RL to optimize energy consumption of buildings by controlling heating, cooling and ventilation systems

  • link to introductory post
  • link to first pilot project results
  • Deployed projects (examples):
    • Construction company headquarters in Copenhagen, Denmark. Control of ventilation rate and air temperature. Running since August 2020.
    • French cosmectics company, office+lab building. Control of air renewal and temperature. Running since October 2021.
    • New university campus in Luxembourg. Control of ventilation in amphitheaters. Running since December 2021.

• Microsoft Bonsai teams up with PepsiCo to create better cheetos

  • link to introductory post
  • Deployment
    • article

• Experience personalization

  • Microsoft personalizer helps provide more personalized recomendations for users.

• AMazon uses RL for supply chain management

  • Supply chain management

• Champion-level drone racing using deep reinforcement learning

  • Nature paper