Kuavo REAL-I: Industrial Embodied AI Data for ICRA 2026 Challenge

LejuRobotics released this industrial dataset for the 1st Real-World Embodied-AI Learning Challenge (REAL-I) at ICRA 2026. It consists of approximately 1,000 image records captured from both simulated and real-world robotic platforms.

Use Cases

• Training vision-based navigation or manipulation models using the image modality
• Testing domain randomization and sim-to-real transfer using the paired simulated and real-world data
• Participating in the ICRA 2026 REAL-I benchmarking challenge

Strengths

• Official dataset for the ICRA 2026 REAL-I competition
• Includes data from both simulated and real-world industrial platforms
• Uses the standard imagefolder format for compatibility with the datasets library

Limitations

• Small sample size of approximately 1,000 records
• Absence of detailed column headers or telemetry metadata in the current release

Provenance

Source

LejuRobotics

Collection Method

Sensor data from real robots and synthetic data from simulation platforms

Freshness

Updated in early 2026 for the ICRA competition cycle.