Security Robots at Train Stations: Urban Hub Deployment

Security Robots at Train Stations: Why Urban Hubs Require a Different Approach

Category-1 train stations in Germany process between 150,000 and 450,000 passengers daily. Classical patrol rounds cover at most 18 percent of relevant floor area per hour at that frequency. [Source required] The gap cannot be closed by adding staff, because the labour market does not supply it.

Three risk levels overlap spatially. On the platform, suicide prevention and track falls are the primary concern. The concourse level is defined by pickpocketing and vandalism. The forecourt brings drug scenes, begging rings, and occasional demonstrations. A uniform patrol logic fails against these profiles.

The night window between 01:00 and 04:30 is systematically understaffed. Incidents during this period account for approximately 41 percent of all operational disruptions on the following day [Source required], ranging from cleaning operations to individual lift closures.

Federal Police (Bundespolizei) and DB Sicherheit operate in parallel with separate jurisdictions. A robot in the station environment must be able to report to both control centres without mixing competencies. Push notifications to the Bundespolizei must be event-based only, not a continuous data stream.

Public space means data protection obligations under GDPR Art. 6(1)(e) in conjunction with the applicable state police law. Recording is the exception; live monitoring is the rule. This principle shapes every robot configuration at a train station.

Next step: TCO comparison classical Wachschutz for quantifying the gap.

Deployment Profile QR-2 in 24/7 Station Operations

The QR-2 patrols concourse levels and covered platforms using a thermal sensor and person detection. Under the RaaS model, the monthly fee is €3,500, including maintenance, updates, and battery replacement. No CapEx applies.

Route length per shift: 14–22 kilometres. 70 percent of the route is defined as a reproducible standard round. 30 percent is triggered on an event-driven basis through control-room override. This separation allows reliable KPI evaluation and response to live situations.

Acoustic detection identifies breaking glass, screams, and noise peaks above 95 dB within 1.8 seconds, and reports position plus audio classification to the control room. False alarms from public-address announcements or traction-unit noise are filtered out via frequency filters.

The thermal image distinguishes prone persons from luggage even at ambient temperatures below 5 degrees Celsius. This function is critical in winter operations. Checks on homeless persons are among the most frequent deployment triggers in covered platform areas, and visual recognition alone fails with sleeping bags.

Battery changes are handled via docking without staff contact. Downtime per cycle is under 12 minutes. With two docking stations per station, availability remains above 23 hours per day.

Platform details: QR-2 for 24/7 outdoor areas.

Platform, Tunnel, Forecourt: Three Zones, Three Sensor Profiles

On the platform, the platform edge functions as a virtual barrier line. When a person approaches within 80 cm of the edge, an acoustic warning is triggered. When a person stands outside the waiting area or a track fall is suspected, an automatic alert is sent to the signal box. The interface must be agreed with the operator; it is not universally predefined.

In tunnels and concourse levels, the LiDAR variant is the recommended choice. Reflections on smooth floor surfaces and low ceilings systematically interfere with camera-based detection. LiDAR delivers reliable distance and movement data regardless of lighting conditions. See QR-3 with LiDAR for tunnels and concourse levels.

On the forecourt, the QR-2 with weather-protection kit IP65 is deployed. The focus is on aggregation points: taxi rank, North entrance, bicycle storage. These points concentrate roughly two-thirds of all external incidents at German urban train stations. [Source required]

Handover zones between robots and fixed cameras require overlapping fields of view of at least 4 metres. Without this overlap, blind spots form at route endpoints and are regularly exploited as hiding places. Planning is conducted on-site, not from floor plans.

Sensor profiles are switched via the control-room dashboard by time of day, not per device. Day profile 06:00–22:00 focuses on person detection; night profile 22:00–06:00 applies increased thermal weighting and a reduced audio threshold. This logic simplifies training and audit.

Data Protection and Property Rights in Public Transport Space

Train stations are classified as mixed spaces under BGH case law. The operator's property rights apply, but the public-purpose designation restricts them. In practice: control measures must be justifiable on transport-related grounds. A blanket filming prohibition inside a station is not enforceable; conversely, full-area biometric recognition is not permissible.

Robot cameras must inform persons under GDPR Art. 13 via visible notices. Pictograms on the device alone are insufficient. Static notice signs at entrances plus a QR code linking to detailed information are required. The Quarero standard template is available in German, English, and French.

Recording is event-triggered only. Continuous recording violates the necessity principle and, based on the practice of state data protection authorities, creates a fine risk above €50,000 per incident. [Source required] The live stream at the control room is unaffected, because it does not constitute storage.

Facial recognition must be excluded from regular operations. Person detection operates at silhouette and movement level. This distinction must be explicitly documented in the data protection concept. It secures acceptance from data protection officers and works councils.

The Data Protection Impact Assessment (DPIA) before the pilot launch is mandatory. A preliminary conversation with the competent state data protection authority (Landesdatenschutzbeauftragter) before contract signing is recommended. This shortens subsequent approval rounds by six to eight weeks.

TCO: Robot Patrol Versus Classical Streife

A 24/7 guard post in German cities costs between €15,000 and €25,000 per month. This figure includes shift supplements, holiday cover, sick-leave absence, and standby reserve. Anyone calculating with €12,000 is excluding the reserve and will encounter problems at the first sick day.

The QR-2 under the RaaS model costs €3,500 per month. No CapEx, no collective-agreement binding, 48-hour delivery time from approval. Maintenance, updates, and battery replacement are included. Details at Robotics-as-a-Service model.

The hybrid configuration of two QR-2 units plus one human intervention post replaces three classical patrol teams at approximately 58 percent lower total cost. [Source required] The intervention post remains necessary, because physical intervention and §34a authority cannot be delegated to machines.

The BDSW documents staff shortages and wage increases in the German security industry. Staff shortage in 2024 stands at approximately 12 percent; wage increase in the collective bargaining year was 7.4 percent. The gap between personnel costs and RaaS fees widens annually, not as a one-off event.

Minimum contract duration is 24 months, after which termination is possible on a monthly basis. Sensor upgrades are delivered without hardware replacement. This clause matters in the transport sector, because regulatory requirements demonstrably increase through 2027.

Integration with DB Sicherheit, Bundespolizei, and the Control Room

Robots report via a VPN-separated channel to the operator's control-room system. The Bundespolizei receives event-based push notifications through a defined interface, not via continuous data stream. This separation protects the jurisdictional boundary and avoids the appearance of auxiliary policing (Anscheinsamtshilfe).

Technically, the system uses ONVIF Profile S for video streaming, MQTT for event data, and REST API for route planning. No proprietary lock-in. Existing control-room systems from Genetec, Milestone, or qognify can be connected without hardware replacement.

The escalation matrix has four tiers. Tier 1: notification in the control-room dashboard. Tier 2: acoustic address by the robot via pre-approved audio tracks. Tier 3: alarm to the intervention team with position data. Tier 4: emergency call 110 via the control-room operator, never by the robot itself.

The shift handover report is generated automatically. Incidents, route kilometres, anomalies, and alarms are available as a PDF. Manual guard logs are eliminated. Legal recognition as documentation evidence is ensured through the GDPR-compliant storage format.

The pilot phase typically runs 90 days with weekly reviews between the security management, works council, and Quarero. Works council participation is not optional. BetrVG §87(1) No. 6 establishes the co-determination obligation.

Legal Framework: NIS-2, KRITIS, and the Transport Sector

Transport is its own KRITIS sector. The framework law defines thresholds and operator obligations in Bundestag-Drucksache 20/9262. Category-1 stations fall under the obligations above a threshold of 500,000 passengers annually. [Source: BSI KRITIS-Verordnung required] Overview of sector logic at KRITIS sectors overview.

NIS-2 requires resilience measures for physical infrastructure in essential sectors, including transport. The common reduction of NIS-2 to IT security is incorrect. Article 21 explicitly names physical protection measures as a component of risk management.

EN ISO 13482 defines safety requirements for personal care robots in public spaces. The QR series is certified to this standard. In procurement procedures, this certification is a required proof, not an optional advantage.

EU Machinery Regulation 2023/1230 replaces the previous Machinery Directive as of January 2027. CE conformity will need to be demonstrated under the new scheme. Quarero manages the transition within the maintenance contract; the operator does not need to organise a new certification process.

Board-level liability applies in the event of failure to implement protective measures. The robot patrol documentation is usable as evidence, because it is maintained without gaps and in a tamper-proof format. Further detail at NIS-2 board liability.

Pilot Operations: 90 Days from Enquiry to Regular Service

Weeks 1–2 are used for site inspection, route definition, and coordination with the data protection officer and works council. The DPIA is produced during this phase and submitted to the state data protection authority. Skipping this step delays the pilot start.

Week 3: delivery of the QR-2 within 48 hours of approval, initial learning runs with a Quarero technician on-site. Routes are driven in both directions and validated at different times of day. Acoustic profiles are calibrated to the specific ambient noise of the station.

Weeks 4–8 are shadow operations running parallel to the classical Streife. KPI collection: detection rate, false alarms, response time, route coverage. Shadow operations allow a reliable comparison and address typical objections raised by works councils.

Weeks 9–12: transition to regular operations, training of control-room staff, review with the Bundespolizei liaison officer. Training covers 8 hours per operator and includes operation of the escalation matrix.

Success criteria are fixed before the pilot starts: response time under 90 seconds, false-alarm rate under 3 percent, coverage above 85 percent of defined floor area per hour. If these values are not met, a clause for contract adjustment or termination applies. No residual-term costs arise.

For connection to adjacent protection concepts, see Perimeter protection landing.

For concrete pilot planning at a specific site: Request a pilot plan, Quarero deployment planning.