Information Gain Regulation in Reinforcement Learning with the Digital Twins' Level of Realism In RlLib
summary
Describes an architecture that uses Reinforcement Learning in simulation together with a Digital Twin to train radio access control policies that minimize radio resource use while keeping robot-cell production KPIs. Shows that using URLLC only ~30% of the time can still achieve accurate real-world robot control, and explains how this concept is implemented using RLlib.
AI-written summary · about
chapters (17)
- 0:00Start
- 0:27Introduction
- 1:19Industry 3.0 for robot control
- 2:23Industry 4.0 for robot control
- 3:01Future factory for robot control
- 3:38Network effects
- 5:13Overview of the automatic QoC setup process
- 5:49Reinforcement Learning in general
- 6:38Existing environments
- 8:05Custom environment-Digital Twin
- 11:12Custom RL algorithm
- 15:41Overview of RLLib
- 18:01Setting up the RLLib Trainer
- 18:18Choosing environment
- 18:39Tagging data with source environment
- 18:59Choosing loss function
- 19:31Conclusions
related talks
56mA Tutorial on Reinforcement Learning II
48mOffline Reinforcement Learning With Only Realizability
30mOffline RL with RLlib
21mApplying Ray and RLlib to Real-life Industrial Use Cases
05mPractical insights into deep reinforcement learning
34mOffline Reinforcement Learning and Model-Based Optimization
49mReinforcement Learning: Demystifying the hype
18mMaximizing Luck in Reinforcement Learning