TartanRGBT: Hardware-Synchronized Stereo RGB and Thermal Images for Robotics

TartanRGBT is a hardware-synchronized robotics dataset from Carnegie Mellon University's AirLab AnyThermal project. It contains 5,952 timesteps of co-registered stereo RGB and thermal imagery across indoor, urban, park, and off-road environments. The dataset was published by Voxel51 in May 2026.

Use Cases

• Train multimodal sensor fusion models based on co-registered RGB and thermal images.
• Benchmark visual odometry and SLAM algorithms based on synchronized stereo trajectories.
• Develop robust object detection for autonomous systems based on imagery from diverse environments.
• Research thermal imaging applications for robotics based on hardware-synchronized data collection.

Strengths

• Contains 5,952 timesteps of data sampled at 1 Hz.
• Features hardware-synchronized, co-registered stereo RGB and thermal image pairs.
• Covers four distinct environment types: indoor, urban, park, and off-road.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect geographic or environmental bias inherent to the specific trajectories collected.

Provenance

Source

Carnegie Mellon University AirLab's AnyThermal project.

Collection Method

Hardware-synchronized data collection from robotic platforms.

Freshness

Last updated 2026-05-26 20:46:34.

Geography

Trajectories cover indoor, urban, park, and off-road environments.