Skip to content

Comprehensive Guide to MyoChallenge 2024

This blog post is designed to help you quickly understand and participate in MyoChallenge 2024. It lists all the resources you can refer to at each stage of the challenge.

The table of contents to the right explains all the stages you'll go through to participate in MyoChallenge 2024, starting from scratch. Feel free to skip any sections you're already familiar with.

Let’s get started!! ✈


🦾 Understand MyoSuite

MyoSuite Logo

MyoSuite is the foundation of MyoChallenge. It's a platform with a collection of environments/tasks to be solved by musculoskeletal models to understand human dexterity and agility. This is an open-source software (OSS) project, with contributions and participation from researchers, developers, and enthusiasts from across the globe.

You can learn basic information about it from these places to get started:

Do you know gym?

MyoSuite is basically a superset of gymnasium (former gym maintained by OpenAI) with musculoskeletal extensions, and it generally adheres to its usage.

If you're not familiar with the gymnasium, it is highly recommended to get familiar with it first.

Learn Usage with Tutorial

To get hands-on with MyoSuite, follow the tutorials provided in the official documentation. These tutorials cover various aspects of using MyoSuite, from setting up the environment to running simulations. Especially, it’s better to check the section “Using Reinforcement Learning” because that is essentially the same process you’ll follow to check you own policy.

🏆 Understand MyoChallenge 2024

MyoChallenge 2024 is a competition held as part of the NeurIPS 2024, focusing on advancing physiological dexterity and agility in bionic humans.

Tasks

MyoChallenge 2024 has two types of tasks:

Manipulation

Manipulation

Transfer objects between two points, involving a hand-off between a biological arm (myoArm) and a prosthetic arm (MPL; Modular Prosthetic Limb). Object properties and environmental factors are randomized for each attempt to thoroughly evaluate model performance.


Locomotion

Locomotion

Navigate varied terrains using a biological leg (myoLeg) and a prosthetic leg (OSL; Open Source Leg). Challenge difficulty is randomized, potentially altering terrain types and introducing obstacles to test adaptability.


Both tasks aim to develop strategies to enable both the biological limbs simulated by MyoSim and the prosthetic limbs to work together successfully and achieve objectives.

For more detailed descriptions, refer to the official documentation of MyoChallenge 2024.

Timeline

Time Stage Detail
Jul - Aug 2024 Open Stage Release of the models, tasks, and environments.
Sep - ⚠ Nov 17, 2024 Scored Stage Submissions open on the scoring platform.
Dec 10, 2024 Results Announcement @NeurIPS 24
Dec 14, 2024 Workshop @NeurIPS 24

Prepare Accounts and Resources

EvalAI

MyoChallenge uses EvalAI to manage the competition (e.g. model submission, ranking scores). You need to create an account and get credentials to submit your model. The step-by-step tutorial on how you can get credentials on EvalAI is explained below.

Before the Scored stage (Sep-Oct), you can refer to the challenge page for MyoChallenge 2023 to understand how the challenge will be handled on this platform.

Development Environment

MyoSuite only requires a Python environment, but it would be advantageous to prepare a computer or cloud environment with GPU resources to accelerate the training process of the policy. A free GPU resource with limited usage is available on Google Colab.

Google Cloud Credits Available!

We are pleased to announce that Google Cloud Platform is sponsoring MyoChallenge '24! Participants are now eligible to receive Google Cloud credits, providing access to advanced computational resources on Google Colaboratory.

Form/Join a Team

Participating in MyoChallenge 2024 can be done individually or as part of a team. Consider the strategy of your participation based on your skills, resources, and goals.

To form or join a team:

  1. Review the team formation rule:

    Team formation / Eligibility: There is no limit to the team size and the affiliation: participants from both academic and industrial institutions are welcome. Each participant will be allowed to join only one team during registration and this cannot be changed during the competition. [ref]

  2. Network with potential teammates through the official MyoChallenge forums, MyoSuite community, social media, etc

  3. Agree on team structure, roles, and communication methods
  4. Register your team on EvalAI

🤖 Develop Your Solution

This is where your skill comes into play! 🔥

Start with Baseline

Before starting the development process, it is recommended to run the pre-trained baselines provided by MyoSuite to assess its performance. Additionally, this can be used as a foundation for your solution.

You can visit the following official tutorials on Google Colab to learn how to:

  • Start a training script that can reproduce the baseline
  • Tune the reward dictionary
  • Access new attributes (e.g., MPL joint angles) and integrate them into your training
  • Submit your policy model

Tutorials for MyoChallenge 2023 can also be your help.

Enhance the Model

Once you've developed your initial model, the next step is to enhance it to improve performance on the tasks. This process may involve experimenting with different RL algorithms, fine-tuning hyperparameters, engineering new features, shaping the reward function, analyzing and visualizing the results, and etc. These techniques can help optimize your model's performance and provide insights into areas for further improvement.

The development process is iterative. Continuously evaluate your model's performance, identify areas for improvement, and refine your approach. Regular evaluation is essential to understand how well your enhancements are working and whether further adjustments are needed.

You may want to take a look at the strategies of previous MyoChallenge winning teams.

Submit Your Solution

After training your model, you can submit it by following the steps below. You can achieve this via GitHub Actions or manually by building and uploading a Docker container. The EvalAI credential info is required for both options.

  1. Obtain your Auth token on EvalAI
  2. Clone the MyoChallenge submission repository: myochallenge_2024eval
  3. Register your token as a secret in the repo
  4. Edit the existing agent file to use your policy
  5. Test your agent
  6. Trigger the action
  1. Install Docker
  2. Obtain your Auth token on EvalAI
  3. Clone the MyoChallenge submission repository: myochallenge_2024eval
  4. Customize Agent Script
  5. Build a docker container with the agent
  6. Upload the container to the EvalAI docker registry

The Colab below also provides detailed step-by-step instructions for the entire submission process of MyoChallenge 2023.

💡 Tips for Success

Leverage Community Resources

MyoSuite has an active community of researchers and developers worldwide. Engage with the community to exchange ideas, seek advice, and stay updated on the latest developments. When you have any questions about MyoSuite/MyoChallenge, you can ask the community on both GitHub Issues and GitHub Discussions.

Should you encounter any issues or bugs that have not yet been raised, please do not hesitate to submit them as an issue on the GitHub repository. This will help us resolve the problem quickly!

You can join the MyoSuite community on our official Slack channel from here: Slack

Collaborate and Learn

Participate in webinars or Q&A sessions organized by the MyoChallenge team to deepen your understanding. Connecting with others and forming a team can also enhance your experience and foster collaboration.

Stay Updated

Please check the MyoChallenge 2024 website regularly for announcements and updates. We also encourage you to follow Myosuite on social media for the latest updates and tips. Staying informed will ensure you don't miss any important notifications.

  • Follow us on X X Follow
  • Give a Star on GitHub GitHub stars



MyoChallenge 2024 is your opportunity to learn and innovate. We encourage you to leverage the resources provided, engage with the community, and enjoy the learning process. Your unique perspective is valuable, and we're excited to see what you'll achieve!

Best of luck in your MyoChallenge journey 🚀