Branding
Robotics Dataset Collection Pipeline — Unitree Go1
ROS2 multimodal data collection pipeline for the Unitree Go1 robot — capturing synchronized camera, IMU, and telemetry streams for training convolutional and transformer-based machine perception models.
Year :
2025
Industry :
Robotics / AI Research
Client :
Florida Atlantic University
Project Duration :
Ongoing
Problem :
Training robust machine perception models for legged robots requires high-quality, synchronized multimodal sensor data — but collecting it reliably across multiple experiment configurations is technically complex and error-prone.
Solution :
Built a ROS2 pipeline using NumPy, SciPy, and OpenCV that captures synchronized camera, IMU, and telemetry streams from the Unitree Go1 across 5+ experiment configurations. Automated AI validation and Guardrail-style data security checks detect corrupt frames and anomalous signals before they reach training.
Tech Stack :
ROS2 · Python · NumPy · SciPy · OpenCV · Unitree Go1/Go2 · Guardrail-style Validation · Convolutional Models · Transformer-based Perception
Impact :
The pipeline generates clean, structured datasets across 5+ experiment configurations, enabling reliable training of CNNs and transformer-based perception models. Guardrail filtering ensures data quality before it reaches any neural network training loop.
