The 1st Real-world Embodied AI Learning Challenge

Note

This manual is being updated in a regular basis…

Welcome to the official competition manual for the 1st Real-world Embodied AI Learning Challenge! This manual covers all the details about the datasets, baseline codes as well as the simulator used in this competition. As for the official competition rules, important dates and other general info, please consult our official competition website .

The competition repositories consist of two parts: The baseline code (kuavo_data_challenge repository) and the simulator (kuavo-ros-opensource repository). Their features are as follows:

• Simulator is based on MuJoCo, with built-in realistic robotic models, restoring real-physics simulation

• Supports Rosbag to Lerobot parquet data conversion

• Imitation Learning (IL) model training framework (DP, ACT)

• Simulator deployment interface as well as automatic high-precision model grading system

• Real-device verification and its deployment

Process

The competition consists of the following steps:

1. Sign up for the competition through the competition website

2. Download datasets and baseline codes as well as environment setup

3. Train your models, you are free to use any learning based methods

4. Submit your models and beat other competitors in the leaderboard!

Please consider joining our discussion group:

To facilitate discussion and raising/solving problems and concerns, we have organised a Discord Server just for our contestants, please consider clicking the link to accept the invitation!

Quick Links Here:

• Competition Website

• Competition Datasets

• Baseline Code Repository

• Simulator Repository

• Discussion Group (Discord)

Open-sourcing Agreement

This competition is open sourced under the MIT License .