Resources

The RobotCycle DevKit streamlines data loading, synchronisation, and analysis for multimodal robotic perception. Built around PyTorch, it provides robust, sensor-specific dataset classes with integrated timestamping, calibration, and preprocessing utilities.

RobotCycle DevKit

A deep learning first Python SDK.

• Comprehensive data loading and processing
• PyTorch Dataset classes for each sensor
• Built-in synchronisation and timestamping
• Optional auto-rectification and calibration

Semantic Annotations

The dataset provides rich multimodal semantic labeling:

• RGB images labeled with SAM (240 images across 60 scenes)
• LiDAR pseudo-labels via Cylinder3D and DBSCAN
• Semantic taxonomy aligned with SemanticKITTI and nuScenes

Ontology and Knowledge Graphs

A structured OWL2/RDFS ontology captures the relationships between:

• Road agents, vehicles, and infrastructure
• Spatial and temporal relations (hasPath, hasLocation, hasTimestamp)
• Sensor metadata and observation links

This enables multimodal reasoning, scene understanding, and traffic knowledge queries.

Visualisation and Analysis Tools

• Risk & proximity analysis for agent interactions
• Eye-gaze heatmap generation extracting attention patterns
• Traffic flow estimation and trajectory tracking
• Map overlays and infrastructure annotation

For more information about the data collection methodology, sensor calibration, open-source toolkit, and analysis results, please refer to our paper:

The Oxford RobotCycle Project: A Multimodal Urban Cycling Dataset for Assessing the Safety of Vulnerable Road Users (PDF)