RoboSense Egocentric Robot Perception Dataset with 1.4M Annotations

RoboSense is a large-scale multimodal dataset containing over 133,000 synchronized sensor readings from cameras, LiDAR, and fisheye sensors. It provides 1.4 million annotated 3D bounding boxes and IDs to facilitate robot perception and navigation research in crowded, unstructured environments. The dataset was created by suhaisheng0527 and last updated in July 2025.

Use Cases

• Train perception models on synchronized Camera, LiDAR, and Fisheye data for multimodal sensor fusion.
• Develop and benchmark 3D object detection algorithms using the 1.4 million 3D bounding box annotations.
• Simulate and test robot navigation policies in crowded environments using the annotated scene data.
• Research egocentric viewpoint understanding for autonomous systems with the provided sensor streams.

Strengths

• Large scale with over 133,000 synchronized multimodal sensor readings.
• Comprehensive 3D annotation with 1.4 million bounding boxes and IDs.
• Specifically targets challenging crowded and unstructured environments for robotics.

Limitations

• Specific column structure and data formats are not detailed in the provided input.
• The geographic origin and temporal range of the collected data are unknown.
• Potential for class or scenario imbalance inherent in real-world, unstructured environment data.

Provenance

Source

suhaisheng0527 on Hugging Face

Collection Method

Constructed dataset for robot perception, likely from real-world robot sensor data collection.

Freshness

Last updated on July 3, 2025.