Simulation Challenge
Overview
The objective of the Simulation Challenge is to successfully fold a given garment. The types of garments
include long-sleeved tops, short-sleeved tops, long pants, and shorts.
Participants are required to conduct environment setup, model deployment, and practical testing based on
the Lehome simulation platform (which is built upon IsaacLab). The usage of Lehome can be found in github.
GitHub Repository
https://github.com/lehome-official/lehome-challengeFeatures
The Simulation Challenge primarily focuses on the model's generalization capability when handling garments of different types and styles.
Therefore, domain randomization mainly targets variations in garment styles and changes in the initial loading positions. All loaded garments exhibit relatively small deformations, and the keypoint information of the garments can be largely exposed.
Garment Samples Loaded in Simulation Challenge
Important Notes
The Simulation Challenge uses a Docker-based submission format for evaluation. Participants can obtain the official Docker image from the GitHub repository and deploy their own policy within the Docker. The Docker image is then uploaded via Google Forms for official evaluation. Please refer to the GitHub README for more detailed information.
Google FormIn the GitHub repository, for each garment category, we provide 10 garments as training samples and an additional 2 garments as test samples. We encourage the use of a wide range of manipulation methods. To facilitate VLA model training, we provide a simulated teleoperation dataset.
Based on the submitted policies, we will further evaluate the models using the official dataset, which includes a larger set of test samples. Check the video below to understand the evaluation metric used in simulation challenge.
The success rate of each team's model will be published on the website, and the top eight teams will advance to the Real-World Challenge, where they will compete on site in Vienna. As a prize for the Simulation Challenge, each qualifying team will be awarded a LeRobot.
