Colabs

Introduction to EAGERx

The best way to get introduced to EAGERx is to play around with the tutorials that are available. They also contain exercises that address common challenges of robotic reinforcement learning and how to overcome them using EAGERx.

The following introductory tutorials are available:

• Tutorial 1: Getting started

• Tutorial 2: Advanced usage

The solutions are available in here.

In the advanced usage tutorial you will learn a quadruped to walk in circles within four minutes of training.

1. Getting Started

This tutorial covers:

• constructing a Graph and an environment using BaseEnv,

• switching between different Engine,

• performing domain randomization.

2. Advanced Usage

In this notebook, you will learn to use EAGERx to create a gym-compatible environment. This tutorial covers:

• how to initialize a robot (Go 1 Quadruped Robot).

• how to add pre-processing nodes (i.e. low-level controllers).

• how to fine-tune low-level controllers to achieve the desired behavior.

• how to (de)select various sensors to investigate its effect on the learning performance.

Developer Tutorials

Next to the introduction tutorials, a set of developer tutorials is also available:

• Tutorial 1: Environment Creation and Training with EAGERx

• Tutorial 2: Reset and Step

• Tutorial 3: Space and Processors

• Tutorial 4: Nodes and Graph Validity

• Tutorial 5: Adding Engine Support for an Object

• Tutorial 6: Defining a new Object

• Tutorial 7: More Informative Rendering

• Tutorial 8: Reset Routines

The solutions are available in here.

The tutorials cover common challenges of robotic reinforcement learning and how to overcome them using EAGERx. The classic control problem of swinging up an underactuated pendulum is used as an example.

1. Environment Creation and Training

This tutorial covers:

• Creating a Graph with an Object.

• How to use this Graph and a Engine to create an BaseEnv.

• How to train a policy with the BaseEnv.

2. Reset and Step

This tutorial covers:

• Extracting observations in the step

• Resetting states using reset()

• The window argument of the connect() method

• Simulating delays using the delay argument of the connect() method

3. Space and Processors

This tutorial covers:

• How to specify a Space

• Creating a custom Processor

• How to add a Processor

4. Nodes and Graph Validity

This tutorial covers:

• Creating a Node

• Adding a Node to the Graph

• Checking the validity of the Graph

• How to make the Graph valid (DAG)

5. Adding Engine Support for an Object

This tutorial covers:

• Adding an engine-specific implementation to an Object

• Initializing the corresponding Engine

• Train with the newly added engine-specific implementation

6. Defining a new Object

This tutorial covers:

• Defining a new Object

7. More Informative Rendering

• Create a layover Node that augments a raw image sensors

• Connect the layover Node and use it for rendering

• Demonstrate that rendering is agnostic to the selected physics-engine

8. Reset Routines

• Defining the reset routine with a ResetNode

• Reset the Object’s with the reset routine.