emilio-cartoni/real2021_starter_kit

Starting kit for the REAL2021 competition on Robot open-Ended Autonomous Learning

Starting kit for the REAL2021 competition on Robot open-Ended Autonomous Learning.

The REAL Competition is an online challenge, open to everyone, aimed at building a robot controller that autonomously learns to interact with an environment in an open-ended way, without being instructed specifically for a task.

Competition has started on August 23rd and will last up to mid June 2022.

REAL 2021 has been launched during the ICDL 2021 conference (video).

Index

• Install the software
• How to launch the agent
• Make your own submission
• How do I...?
• Support

Install the software

Pre-requisites: Python, Git, Conda (Anaconda or Miniconda) and Docker
All software has been tested on Ubuntu 18.04 and 20.04.

1. Clone this repository:

git clone https://github.com/emilio-cartoni/REAL2021_starter_kit.git  
cd REAL2021_starter_kit

2. Create a conda environment named "real_robots" from the provided environment.yml

conda env create -f environment.yml

This could last a while as many libraries that are needed by the baseline will be installed, including Tensorflow.
(It is possible to reduce the libraries in the environment if you do not want to run the baseline).

How to launch the baseline agent

1. Activate the real_robots conda environment

conda activate real_robots

2. Launch the local evaluation:

python local_evaluation.py

The default values will only launch a short evaluation, edit local_evaluation.py to launch full evaluations and customize the options.

Make your own submission

To make submissions to the challenge, it is necessary to create and submit a docker container, which will contain all the dependencies to run your submission and the simulation.

1. Launch the image build to create a docker image for your submission.

   build.sh
   

   (The first build may take a long time since it will have to recreate the environment inside the container - subsequent build will be faster due to docker caching mechanism)
2. (Optional) You can launch your docker image locally, to check it is working:

   docker run -it real2021submission:latest /root/miniconda3/bin/conda run -n real_robots python docker_evaluation.py
   

3. Go to EvalAI Submit page and follow the instructions there to submit the image as a submission for the REAL2021 competition.

How do I...?

For further information, check our Wiki.
In particular, if you want to know more about:

• the environment, see here
• the options available for the environment and the simulation, see here
• how to develop your own controller and the docker image build, see here
• to know more about the baseline, see here

... or see the FAQ for common questions or errors.

Support

Feel free to contact us for any problems or additional questions on the competition.
You can post them as an issue here on Github, or post on EvalAI forum or, if you prefer, directly contact the organizers via mail.

Best of Luck 🎉 🎉

real2021_starter_kit's Issues

Problems with video saving in local_evaluation.py

Here, I'm encountering another problem.
I'm trying to generate a video while doing experiments by changing the parameters in line 30 of local_evaluation.py
It seems it failes to read the camera image, and I have no idea how to fix it. The whole environment works well if line 30 is removed.
Thanks for any suggestion.
The attachment contains the code I ran and the output log it produces.