Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/armandpl/furuta

Building and Training a Rotary Inverted Pendulum robot
https://github.com/armandpl/furuta

deep-learning reinforcement-learning robotics

Last synced: 3 months ago
JSON representation

Building and Training a Rotary Inverted Pendulum robot

Awesome Lists containing this project

README 

        
# Furuta Pendulum



In this repository you will find everything you need to build and train a rotary inverted pendulum (also known as a furuta pendulum). Watch the build video [here](https://www.youtube.com/watch?v=Y6FVBbqjR40).



![](./assets/pendulum.jpg)

## Table of Content

- [Motivations and Goals](#motivations-and-goals)

- [Getting Started](#getting-started)

- [MLOps](#mlops)

- [Credits](#credits)



## Motivations and Goals

- There are many ([hundreds!](https://www.youtube.com/results?search_query=furuta+pendulum)) videos of rotary inverted pendulum on the internet but we couldn't find good documentation on how to build one and how to control one. This projects aims to make building a furuta pendulum **affordable and reproducible**.

- We wanted to learn about RL. Balancing a pendulum is kind of the "hello world" of the field, it seemed like a good starting point.



## Getting Started

1. [Get the supplies and follow the hardware instructions](https://github.com/Armandpl/furuta/tree/master/hardware)

2. [Follow the software instructions](https://github.com/Armandpl/furuta/tree/master/software)

3. Deploy pre-trained models, run experiments and train the robot



If you have any question feel free to open an issue or DM me [@armand_dpl](twitter.com/armand_dpl).



## Usage

1. Plug-in the robot

2. Run `sudo dmesg | grep tty` in terminal to find which port is used for the device

3. run `python tests/interactive_robot_self_test.py`



## MLOps

During this project we leveraged [Weights and Biases](https://wandb.ai/site) MLOps tools to make our life easier. You can find our experiments, pre-trained models and reports [on our dashboard](https://wandb.ai/armandpl/furuta). You can also read more about [Training Reproducible Robots with W&B here](https://wandb.ai/armandpl/furuta/reports/Training-Reproducible-Robots-with-W-B--VmlldzoxMTY5NTM5).



## Credits

To make this robot work we built on top of existing work!

- We got the encoder precision and the idea to use a direct drive motor from the [Quanser Qube design](https://quanserinc.box.com/shared/static/5wnibclu7rp6xihm7mbxqxincu6dogur.pdf).

- We re-used bits from [Quanser's code](https://git.ias.informatik.tu-darmstadt.de/quanser/clients/-/tree/master/quanser_robots/qube). Notably:

  * their VelocityFilter class to compute the angular speeds

  * their GentlyTerminating wrapper to send a zero command to the robot at the end of each episode

  * their rotary inverted pendulum simulation

  * their ActionLimiter class

- The arm assembly is inspired by this [YouTube video](https://www.youtube.com/watch?v=xowrt6ShdCw) by Mack Tang.

- The visualization we use for the simulation is copy-pasted from https://github.com/angelolovatto/gym-cartpole-swingup

- We use the [StableBaselines3](https://github.com/DLR-RM/stable-baselines3) library to train the robot.

- We implemented tricks from [Antonin Raffin's talk at RLVS 2021](https://www.youtube.com/watch?v=Ikngt0_DXJg).

  * HistoryWrapper and continuity cost

  * [gSDE](https://arxiv.org/abs/2005.05719)

- We use [code from Federico Bolanos](https://github.com/fbolanos/LS7366R/blob/master/LS7366R.py) to read the encoders counters.



## Authors

[Armand du Parc Locmaria](https://armandpl.com)

[Pierre Fabre](https://www.linkedin.com/in/p-fabre/)