Explore an introduction to reinforcement learning in this 30-minute conference talk from ODSC East 2020. Dive into the world of training robots through a subset of machine learning, presented by Craig Buhr, PhD, Engineering Manager at MathWorks. Learn about reinforcement learning applications, control goals, and the differences between traditional and alternative approaches to robot control. Discover key terminology, workflow, and the importance of simulated environments in the reinforcement learning process. Gain insights into defining rewards, creating actor-critic networks, and designing effective agents. Understand the significance of reward function design and shaping for improved learning outcomes. Finally, explore the deployment of trained policies to target hardware platforms, bridging the gap between simulation and real-world applications.
How to Train Your Robot - An Introduction to Reinforcement Learning
Overview
Syllabus
Intro
Reinforcement Learning. A Subset of Machine Learning
Reinforcement Learning Applications
The goal of control
A walking robot - a traditional controls approach
A walking robot-an alternative approach
What is Reinforcement Learning?
Some Reinforcement Learning Terminology
Reinforcement Learning Workflow
Real vs Simulated Environments
Define Simulated Environment
Defining the Reward
Actor-Critic Training Cycle
Creating the Agent
Create Critic Network
Create Actor Network
Defining the Agent
Training our Deep Reinforcement Learning Agent
Training the Agent
Reward Function Design Matters
Reward Shaping to Improve Learning
Deploy policy to the target hardware
Policy Deployment to Hardware Platforms
Taught by
Open Data Science
Related Courses
-
Overview of Deep Reinforcement Learning Methods
-
Task-Driven Network Discovery via Deep Reinforcement Learning on Embedded Spaces
-
TF-Agents - A Flexible Reinforcement Learning Library for TensorFlow
-
Introduction to Reinforcement Learning
-
Challenges in Reward Design for Reinforcement Learning-based Traffic Signal Control - An Investigation Using CO2 Emission Objective
-
TinyRL: Teaching AI to Swing Up a Real Pendulum
