Multi-Terrain Humanoid Locomotion Challenge

Overview

This challenge focuses on developing robust locomotion controllers for humanoid robots to navigate complex, unstructured terrains - a critical capability for real-world deployment in homes, disaster zones, and natural environments.

Using the high-performance Isaac Gym simulator, participants will create locomotion policies that enable humanoids to traverse diverse surfaces including slopes, stairs and rubble terrain. The competition provides standardized robots and terrains, allowing you to focus on advanced locomotion strategies.

Competition Objective

Your goal is twofold:

1. Maximize traversal distance - Go as far as possible on each terrain
2. Minimize completion time - Achieve the same distance as fast as possible

The final score combines both distance coverage and efficiency metrics.

Participation

🏆 Awards

Thanks to our sponsors at Unitree and Fourier: 🥇 First Prize ($1000) 🥈 Second Prize ($500) 🥉 Third Prize ($300)

Participants need to open-source their code to be eligible for awards. However, you can still participate in the challenge without open-sourcing your code. If you want to change your response regarding open-sourcing your code, please contact us.

Evaluation

Test Scenarios

Policies will be evaluated on three specialized scenarios (detailed in Scenarios):

1. Robustness Testing: Extended terrain sequences testing endurance
2. Generalization Testing: Novel terrain combinations testing adaptability
3. Extreme Testing: Maximum-difficulty terrains testing capability limits

Scoring Metric

We compute two metrics for each evaluation episode:

• Completion Rate (higher is better): Percentage of traversed distance over total length
• Efficiency Score (lower is better): Elapsed timesteps normalized by max timesteps (only for completed episodes)

Scoring Formula

The score will be averaged across all scenario episodes for each robot. The award is selected based on the total score summed over all robots.

Codes

Our open-source Terrain Benchmark is accessible in this repository.

Resources

• Terrain Suite: A diverse set of terrains implemented in Isaac Gym, including slopes, stairs, rubble, uneven ground, and slippery surfaces.
• Training Platform: A foundational training platform for developing and testing locomotion control policies.
• Documentation: Comprehensive documentation to guide participants in using the benchmark effectively.

Reference Papers

• One Policy but Many Worlds: A Scalable Unified Policy for Versatile Humanoid Locomotion.
• Learning Humanoid Locomotion with Perceptive Internal Model.
• Humanoid Parkour Learning.
• BeamDojo: Learning Agile Humanoid Locomotion on Sparse Footholds.
• Learning Perceptive Humanoid Locomotion over Challenging Terrain.

Scenarios

We have designed different scenarios to evaluate the different performance of robots. All terrain used for evaluation comes from the single terrain model of the terrain module provided in the code.

1. Robustness Evaluation: To evaluate the robustness of the robot, we will extend the size of the terrain several times. Taking stair terrain as an example, we will expand the 5-10 steps used during training to 20 or even more steps to test the robustness of the robot.
   

   Stair terrain for Robustness evaluation
2. Extreme Evaluation: To assess the upper limits of locomotion strategies, we conduct extreme terrain evaluations. For example, the height of stairs will be set significantly higher than in typical scenarios. These challenging conditions are designed to test the maximal capability of the trained humanoid policies, revealing potential failure modes and identifying areas for further improvement.
   

   Stair terrain for Extreme evaluation
3. Generalization Evaluation: To evaluate the generalization performance of locomotion strategies, we create complex scenarios by combining multiple types of terrain within single environments. By requiring robots to transition smoothly and adaptively between different terrain elements—such as stairs, slopes, rubble, and narrow bridges—we can more rigorously assess their ability to generalize learned skills and robustly handle a variety of unseen or mixed conditions in real-world environments.

Simple Terrain

Normal Terrain

Hard Terrain

Challenging Terrain

Rules

1. Eligibility: Participation is open to academic, industry, and independent teams worldwide.
2. Fair Play: Use of cheating techniques, hard-coding the test environment, or exploiting simulator bugs is strictly prohibited.
3. Final Decisions: The organizing committee reserves the right to make final decisions regarding rule interpretation and winner selection.

Organizers

Ye Shi

ShanghaiTech University

Tianxiang Gui

ShanghaiTech University

Yahao Fan

ShanghaiTech University

Shutong Ding

ShanghaiTech University

Li Zhang

Fudan University

Xin Jin

Eastern Institute of Technology, Ningbo

Jiayuan Gu

ShanghaiTech University

Jingya Wang

ShanghaiTech University