commaSteeringControl is a public dataset released by comma.ai containing over 12,000 hours of real-world driving data recorded while the openpilot autonomous driving system was active. It includes detailed telemetry related to vehicle steering control, such as steering angle, steering torque, lateral acceleration, road roll, vehicle speed, and discrepancies between desired and actual movement.
This dataset is particularly valuable for researchers and engineers working on Advanced Driver Assistance Systems (ADAS), autonomous vehicles, and vehicle dynamics modeling. It can be used to train machine learning models for lateral control (e.g. steering prediction), simulate vehicle behavior, or fine-tune stability and traction control systems. With a wide range of car models and driving conditions, the dataset supports the development of robust and generalizable control algorithms.
Use case:
Lateral Control Modeling for Autonomous Driving Systems
| Example Observation ๐ โ | Front camera images (20 Hz, 320ร192) Vehicle speed (m/s) Longitudinal & lateral acceleration Yaw rate, pitch, roll Lane position (offset, heading) Turn signal state, regenerative braking |
| Example Action ๐โ | steering_torque_cmd (โ1 to +1) throttle_cmd (0โ1) brake_cmd (0โ1) |
| Example Reward | r = โ wโยท |
| Example Transitions | Recording at 20 Hz for โ12,500 h โ 12,500 ร 3,600 ร 20 โ 9 ร 10โธ pairs (sโ, aโ, sโโโ) |