LejuRobotics/kuavo_data_challenge_icra

[ICRA 2026] THE 1ST REAL-WORLD EMBODIED-AI LEARNING CHALLENGE DATASETS

✨ Overview

The 1st Real-World Embodied-AI Learning Challenge (REAL-I) is a competition held at ICRA 2026 and organized by LEJU ROBOT. We provide:

• open access to real robots for evaluation and benchmarking
• a large-scale industrial dataset collected from both simulated and real platforms
• prizes and community support for top submissions

Visit the Official Website for more information and get involved.

🤖 Hardware Platform

The main hardware platform is Kuavo 4 Pro with the following features:

• Robot parameters: Height 1.66 m, weight 55 kg, supports hot-swappable batteries
• Motion control: 40 degrees of freedom, max walking speed 7 km/h, supports bipedal autonomous SLAM
• Generalization: Supports multi-modal large models (e.g., Pangu, DeepSeek, ChatGPT), with 20+ atomic skills

📦 Dataset summary

This dataset contains both simulated and real robot data. Each task includes 1,000 episodes. The structure is as follows:

kuavo_data_challenge_icra
└── sim
    ├── TASK1-ToySorting
    ├── TASK2-ParcelWeighing
    └── TASK3-ConveyorBeltSorting
|
└── real(not updated yet)
    ├── TASK1-RubbishSorting
    ├── TASK2-ParcelWeighing
    └── TASK3-ConveyorBeltSorting

For maximum preservation of the valuable information contained in the original data, all of this competition’s datasets are in rosbag format. Competitors can freely use the abundant sources of sensor data inside the rosbag files for their model training.

See the Competition Manual for complete dataset specifications.

🎯 Competition Tasks

The competition consists of three carefully designed tasks, each available in both simulation and real-machine tracks. These tasks are representative of typical industrial applications and robotic workflows.

Simulation track

Task 1: Toy Sorting

Assorted toys are randomly laid out on a desk. The robot must grasp and sort them: place animal toys in the right basket and car toys in the left basket. The robot starts from a randomized position away from the desk; desk height and toy arrangement vary.

• Scoring (100 pts max):
  • 40 pts per correct toy placement
  • 20 pts for completion within timeframe (1-pt penalty per extra second)

Task 2: Parcel Weighing

The robot picks up a soft-pouch parcel from a moving conveyor belt, places it on an electronic scale for weighing, then transfers it to another conveyor belt. Scale and parcel positioning are randomized.

• Scoring (100 pts max):
  • 40 pts for correct placement onto the scale
  • 40 pts for correct placement after weighing
  • 20 pts for completion within timeframe (1-pt penalty per extra second)

Task 3: Conveyor Belt Parts Sorting

The robot picks up industrial components in random orientations from a moving conveyor belt and places them in the correct sorting bins. Four components must be sorted successfully.

• Scoring (100 pts max):
  • 20 pts per successful grasp and placement of each component
  • 10 pts for completing all four components in sequence
  • 10 pts for completion within timeframe (1-pt penalty per extra second)

Real-machine track

Task 1: Rubbish Sorting

Assorted refuse is randomly laid out on a desk. The robot must grasp and sort: place recyclable items in the blue bin and other rubbish in the grey bin.

Task 2: Parcel Weighing

Identical to the simulation version: pick up a parcel from a moving conveyor, weigh it on an electronic scale, and transfer to another belt.

Task 3: Conveyor Belt Parts Sorting

Identical to the simulation version: pick up industrial components from a moving conveyor in random orientations and sort them into correct bins (four components total).

For complete task specifications, see the Competition Manual.

👥 Community

If you are participating (or planning to join) the competition, you are welcome to join our Discord channel to discuss with other participants, share insights, and stay updated.