BMW says it is expanding a consent-based programme to collect event-triggered image, environmental-sensor and driving-dynamics data from suitably equipped customer vehicles in EU member states. The company says the data will be used to improve driver-assistance systems and partially automated-driving functions through machine learning.
The announcement, published on 7 September 2026, describes a manufacturer data-collection programme rather than an independently evaluated safety study. BMW does not provide measured improvements in collision avoidance, driver assistance or road safety.
Contents
What BMW is collecting
BMW says collection from customer vehicles in EU member states began in April 2026. It applies to vehicles from the company’s new model generation that have the required sensor technology and data-processing architecture.
The press release lists the new BMW iX3 and i3, X5 and 7 Series, as well as subsequent new models and model updates, among the vehicles covered. Participation depends on customer consent.
The data may include:
- Images from exterior cameras showing the vehicle’s surroundings;
- Environmental-sensor data;
- Driving-dynamics information, including speed, direction of travel and steering angle.
These sources provide different views of the same driving situation. Cameras can record visual features such as lane markings, vehicles and road users, while other environmental sensors and vehicle-motion data help describe how the car and its surroundings were changing. Driver-assistance systems commonly combine these inputs to support functions such as braking, lane-related assistance and collision avoidance.
BMW says the collection is intended to support the continuous development of driver-assistance systems and the further development of semi-automated-driving functions. It also says that future software updates may deliver improvements based on insights from real-world traffic data.
How event-based collection works
BMW describes the programme as event-based rather than as continuous transmission of all vehicle-camera footage. In this approach, a defined driving situation or system event selects data for collection.
The examples in BMW’s announcement include:
- A prevented collision during a highway lane change;
- Activation of emergency braking;
- Heavy manual braking by the driver;
- A sudden evasive manoeuvre.
Event-based collection can reduce the amount of data transmitted compared with sending all continuously captured footage. It can also focus development teams on situations in which assistance systems have had to interpret a complex or potentially dangerous combination of road, vehicle and driver behaviour.
BMW says event-based video data may be collected in unaltered form for a maximum of 120 seconds per event, with this form of video collection beginning in mid-September 2026. The source does not specify the exact event-detection logic, whether every listed trigger produces a recording, or whether all associated sensor and driving-dynamics data are collected for the same 120-second period.
The company says unaltered image data is used through machine learning to develop driver-assistance and semi-automated-driving functions. Machine-learning development commonly relies on selected or labelled real-world examples to assess and improve a system’s perception and decision-making. BMW’s announcement, however, does not describe its specific labelling process, model architecture, training pipeline or evaluation results.
BMW’s stated privacy controls
BMW says collection and use comply with applicable data-protection regulations in EU member states and that participation is subject to customer consent.
The company describes several privacy-by-design measures:
- The vehicle identification number is deleted immediately after transmission to BMW’s IT backend. BMW says the data can then no longer be linked to a specific vehicle.
- BMW says its systems do not identify individual road users.
- If authorised BMW employees need to access individual recordings for development, the faces and licence plates of other road users are obscured before playback to the extent technically possible.
These measures address different parts of the data-handling process. Removing the vehicle identification number is intended to prevent the collected material from remaining linked to a specific vehicle. Masking faces and licence plates before authorised employee playback is intended to reduce exposure of identifiable details during human access.
BMW’s statement does not establish that original imagery is blurred before transmission or initial storage. It also does not specify the data’s retention period, storage location, access-control arrangements or deletion process beyond the stated removal of the vehicle identification number after transmission. The announcement does not explain the detailed consent interface, whether consent can be withdrawn, or what happens to a vehicle’s participation after withdrawal.