🏁 Welcome to the AI for Prosthetics Challenge

Welcome to the AI for Prosthetics Challenge, one of the official challenges in the NeurIPS 2018 Competition Track. In this competition, you are tasked with developing a controller to enable a physiologically-based human model with a prosthetic leg to walk and run.

You are provided with:

• A human musculoskeletal model.

• A physics-based simulation environment, OpenSim, to synthesise physically and physiologically accurate motion.

• Datasets of normal gait kinematics.

Your agent will be scored based on how well it adapts to a requested velocity vector changing in real time.

➡️ Follow the instructions on our GitHub repository to get started!

🎯 Our Objectives

• Bring Deep Reinforcement Learning to solve problems in medicine.

• Promote open-source tools in RL research (OpenSim, RL environment, and the competition platform are all open-source).

• Encourage RL research in computationally complex environments featuring stochasticity and high-dimensional action spaces.

🔥 Updates

Update 11/02:
For urgent issues, please visit the Gitter channel and the Round 2 GitHub issue.

Update 11/01:
Google Cloud Platform will generously sponsor the second round of the challenge. The top 50 teams from Round 1 will be awarded $400 cloud credits!

Update 10/31:
Instructions for submission for Round 2 are available here.

Update 07/30:
Watch our webinar to learn more about biomechanics, neuroscience, and reinforcement learning!

🆚 What’s New Compared to NIPS 2017: Learning to Run?

Several improvements have been made based on feedback from the last challenge:

• Experimental data can now be used to speed up the learning process.

• The third dimension has been introduced in the OpenSim model (models can fall sideways).

• A prosthetic leg has been added to address medical challenges in prosthesis modelling.

➡️ Not familiar with NIPS 2017: Learning to Run? Watch this video!

📏 Evaluation Criteria

You are tasked with building a function f which:

• Takes the current state observation (a dictionary describing the current state).

• Returns the muscle excitations action (a 19-dimensional vector) to maximise the total reward.

The trial ends when:

• The pelvis of the model falls below 0.6 metres, or

• You reach 1000 iterations (equivalent to 10 seconds in the virtual environment).

🥇 Round Details

Round 1

• Reward formula: 9 * s - p * p

  • s = number of steps before stopping.

  • p = absolute difference between horizontal velocity and 3.

• Objective: Run at a constant speed of 3 metres per second.

Round 2

• Task: Follow a requested velocity vector that changes over time based on a random process.

• Distribution of this process will be provided in mid-July.

📅 Timeline

• Round 1: 16.06.2018 – 28.10.2018

• Round 2 (Test Run): 1.10.2018 – 28.10.2018

• Round 2: 29.10.2018 – 6.11.2018

➡️ Top 50 teams from Round 1 qualify for Round 2.

📚 Resources

Visit the osim-rl project website for resources on biomechanics, reinforcement learning, and NeurIPS 2017 solutions. Also visit the OpenSim website for materials on musculoskeletal simulations.

Interesting Blog Posts by Participants

• Understanding the Challenge

• Understanding the Action Space

• Understanding the Observation Space

• Understanding the Reward

Helper Libraries

• osim-rl-helper by seungjaeryanlee

Additional Reading

• Our NIPS 2017 Learning to Run Approach

• Research Paper on NIPS 2017 Challenge

• Run Skeleton Run - 3rd Place Solution

📣 Contact Us

For queries or technical issues, use the following public channels:

• Gitter Channel: crowdAI/NIPS-Learning-To-Run-Challenge

• Technical Issues: GitHub Issues

• Discussion Forum: CrowdAI Challenge Discussion

For private communications (only if absolutely necessary):

• Email: lukasz.kidzinski@stanford.edu

• Email: sharada.mohanty@epfl.ch

🏆 Prizes

Prizes confirmed so far:

• 1st Place: 2 × NVIDIA Titan V

• 2nd Place: 1 × NVIDIA Titan V

• 3rd Place: 1 × NVIDIA Titan V

Additional prizes:

• Invitation to publish articles in the NeurIPS competition book.

• Invitation to the 3rd Applied Machine Learning Days at EPFL, Switzerland (January 26–29, 2019), with travel and accommodation covered.

• Invitation to give a research talk at Stanford, covering travel and accommodation.

• Reimbursement of travel and accommodation for NeurIPS 2018.

Media