Cameras on Outdoor Robots: Privacy, GDPR and Responsible Data Collection
Tom Cosgrove,

TL;DR: Outdoor and workplace robots with cameras inevitably record people, which turns robot privacy into a practical design and research question. In Europe, images of identifiable people are personal data under the GDPR, so principles such as purpose limitation, data minimization, storage limitation and security apply. Teams can reduce risk by collecting only what the task needs, anonymizing early, limiting access and documenting decisions. This article is general guidance, not legal advice.
Cameras are among the most useful sensors on a mobile robot. They support navigation, inspection, mapping and machine learning. The same cameras, however, capture passers-by, employees, vehicles and sometimes the inside of private spaces. Research teams and integrators who plan robot deployments in public areas, campuses, farms or industrial sites need a clear approach to this data before the first field test, not after.
Why is robot privacy a real concern for camera-equipped mobile robots?
Robot privacy becomes an issue whenever a robot’s sensors can identify people or reveal information about them. Unlike a fixed security camera, a mobile robot moves through spaces, records from many angles and often stores long, continuous logs for later analysis. That combination can create detailed records of who was where and when.
Situations in which robots commonly capture personal data include:
- field tests on campuses, streets, parks or other publicly accessible areas,
- inspection and monitoring in workplaces where employees are present,
- agricultural work near farm workers, neighbors or public paths,
- dataset collection for machine learning, where raw images are stored for a long time,
- remote operation, where live video is streamed to operators.
Beyond legal requirements, robot privacy is also about trust. People who feel watched by a robot are less likely to accept it, and a privacy incident can damage a research group’s or a company’s reputation and its access to test sites.
What do GDPR principles mean for robot privacy in general terms?
The GDPR applies to the processing of personal data, and images or video in which a person can be identified generally count as personal data. Other data can also be personal, such as precise location traces tied to an individual or audio recordings of voices. The regulation sets out several principles that translate naturally into robot privacy practices:
- Lawfulness, fairness and transparency: there must be a valid legal basis for recording, and people should be informed that a robot is collecting data.
- Purpose limitation: data collected for navigation research should not quietly be reused for unrelated purposes.
- Data minimization: collect only the data needed for the stated purpose, at the resolution and duration required.
- Accuracy: keep data correct where it relates to individuals.
- Storage limitation: delete or anonymize data once it is no longer needed.
- Integrity and confidentiality: protect data against unauthorized access, loss or leaks.
- Accountability: be able to show how these principles are applied.
Systematic monitoring of publicly accessible areas or large-scale processing may also require a formal data protection impact assessment. Rules differ by country and by type of organization, and other regions have their own privacy laws, so specific obligations should always be confirmed with the institution’s data protection officer, legal team or ethics board.
How can teams apply data minimization on a robot?
Data minimization is the most effective robot privacy measure because data that is never recorded cannot be misused or leaked. On a mobile robot it can be applied at several points in the pipeline:
- Sensor choice: where the task allows, use sensors that do not identify people, such as LiDAR or depth-only data, instead of full-color video.
- Field of view: point cameras at the ground, crops or assets rather than at head height, and mask areas that are not relevant.
- Resolution and frame rate: record at the minimum level the analysis needs.
- Selective recording: log camera data only during specific tasks or in defined zones, and avoid continuous recording by default.
- Onboard processing: extract the needed information on the robot and store results, not raw images.
Processing on the robot itself, a form of edge computing, is especially useful because raw images never need to leave the device. Only detections, measurements or anonymized frames are transmitted or stored.
What role does anonymization play in robot privacy?
Anonymization removes or obscures information that identifies individuals. In camera data this usually means blurring or masking faces and license plates, and sometimes whole bodies. Applied early in the pipeline, anonymization lets teams keep useful visual data while lowering privacy risk.
Several points are worth keeping in mind:
- Anonymize as early as possible, ideally on the robot before data is stored or transmitted.
- Automatic detection is imperfect, so people at the edge of frames, partially hidden or far away may be missed. Review samples manually.
- Context can still identify people, for example a distinctive uniform combined with a location and time.
- Pseudonymized data is still personal data if it can be linked back to individuals with additional information.
- Keep originals only when necessary, with stricter access and a clear deletion date.
Anonymization also affects research quality. Blurred regions can influence machine learning models, so teams training perception models should consider how masking interacts with their task.
How should research teams store and share robot recordings?
Field recordings are valuable research assets. They allow algorithms to be replayed, compared and published. They also accumulate personal data quickly. Good robot privacy practices therefore extend from the robot to the lab’s storage and sharing workflows. The article on recording and replaying field data with ROS 2 bags describes the research side of this workflow, and the same recordings need privacy-aware handling.
Practical measures include:
- encrypting storage on the robot and on lab servers,
- limiting access to the people who need the data and logging that access,
- separating raw recordings from anonymized, shareable versions,
- defining retention periods and deleting data when they expire,
- reviewing datasets before publication or sharing with partners,
- documenting where data came from and which consent or legal basis applies.
When recordings become training data, these rules should carry through the whole pipeline. The guide on building an ML training pipeline from field recordings is a natural place to add anonymization and retention steps.
What practical robot privacy policies work for field tests?
A short written policy, agreed before field work begins, makes robot privacy manageable for students, engineers and partners alike. Useful elements of such a policy include:
- Purpose statement: what the data is for and what it will not be used for.
- Data inventory: which sensors record what, where and for how long.
- Transparency measures: signs at test sites, information for site owners and employees, and a contact point for questions.
- Minimization and anonymization rules: default settings for cameras and processing.
- Access and retention: who can see raw data, and when it is deleted.
- Incident handling: what to do if data is lost or shared by mistake.
- Review: approval by the data protection officer or ethics board where required.
Open platforms make it easier to put such policies into practice because teams can control exactly which sensors are mounted, what is recorded and where processing happens. Fictionlab’s robots run open-source software, which gives research groups that level of control. More on how labs use mobile robots in studies is available on the research applications page.
FAQ: robot privacy and responsible data collection
Are robot camera images personal data under the GDPR?
Generally yes, when a person can be identified directly or indirectly from them. Images in which no person can be identified are usually not personal data, but this needs to be assessed case by case.
Does a research purpose remove GDPR obligations?
No. Research can benefit from specific provisions and exemptions, but core principles such as security, minimization and transparency still apply. Institutions usually have their own procedures for research data.
Is blurring faces enough to make data anonymous?
Not always. Other features, locations and timestamps can still identify people. Anonymization should be evaluated in context, and in doubt the data should be treated as personal.
Do LiDAR point clouds raise robot privacy concerns?
Usually far fewer than camera images, because they rarely identify individuals. Combined with other data, such as location and time, they can still reveal information about people, so they deserve some care.
Should people be informed when a robot records in a public space?
Transparency is a core GDPR principle, so informing people, for example with clear signs and contact information, is generally expected. The exact requirements depend on the setting and local rules.
Who should approve a robot data collection plan?
Typically the institution’s data protection officer, legal team or research ethics board, depending on the organization. Involving them early avoids delays before field tests.
If your project needs a robot designed around specific sensors, onboard processing or privacy requirements, explore Fictionlab’s custom robotics services.