Purpose

GroundCloud’s AI-enabled Video Event Data Recorder (VEDR) uses advanced camera systems and machine learning to detect unsafe driving behaviors inside and outside the vehicle. This article explains how the AI works, what types of events it detects and what drivers can expect to see or hear in the cab.

MV/AI Overview

MV/AI VEDR (Machine Vision / Artificial Intelligence, referred to in this article simply as AI) combines:

• Cameras (road-facing and driver-facing).

• Sensors (accelerometer and GPS).

• AI models trained to recognize behaviors and risk patterns.

The system continuously analyzes road conditions, vehicle movement and driver behavior. When it detects risky or unsafe activity, it may issue audible warnings in real time and may report violations for human review.

AI helps the system identify patterns that traditional accelerometer-only devices cannot detect, enabling more effective safety monitoring and greater awareness of high-risk behaviors.

AI Safety Systems

GroundCloud’s AI-enabled VEDR system uses three key technologies to monitor safety: Driver Monitoring System (DMS), Advanced Driver Assistance System (ADAS) and Driver Safety Assistance (DSA). Each plays a different role in identifying unsafe behaviors and helping drivers correct them.

• DMS focuses on the driver inside the cab.

• ADAS monitors the road environment outside the vehicle.

• DSA ties these systems together by providing audible, in-cab alerts when DMS or ADAS detects risky behavior.

Together, DMS, ADAS and DSA create a comprehensive, real-time safety system that helps drivers stay aware, avoid risky situations and improve safety performance throughout each trip.

Driver Monitoring System (DMS)

DMS uses the driver-facing camera to detect behaviors inside the vehicle that may indicate distraction, inattention or unsafe practices.

DMS analyzes the driver’s head position, face, posture, seat belt status, and general in-cab behavior. When the system observes a high-risk or prohibited action, it issues audible alerts and may log a violation.

Examples of DMS-detected behaviors include:

• No seat belt

• Using phone

• Distracted driving

• Drowsiness

• No driver detected

Advanced Driver Assistance System (ADAS)

ADAS uses the road-facing camera to detect behaviors and risks outside the vehicle, such as proximity to other vehicles, lane positioning or potential collisions.

ADAS analyzes what is happening on the road ahead, vehicles, lane markings, pedestrians and potential collision risks. It detects many types of risky driving behaviors and road rule violations.

Common road behaviors detected by ADAS include:

• Close following

• Tailgating

• Rolling stop

• Lane departure

• Forward collision

Driver Safety Assistance (DSA)

Driver Safety Assistance (DSA) is the in-cab alerting system built into AI-enabled VEDR devices. Its purpose is to help drivers self-correct unsafe behaviors in real time, reducing risk before a violation is recorded.

DSA issues short, audible alerts whenever the AI detects a behavior that could lead to a violation, such as distraction, tailgating, lane departure or failure to wear a seat belt. These alerts allow the driver to adjust their behavior immediately. If the driver corrects the issue, no violation is logged.

For certain behaviors (for example, distraction or seat belt), DSA typically provides two warnings before escalating to a violation. For urgent or high-risk situations (such as forward collision or failure to stop), there may be no opportunity for a warning and the device will record the violation immediately.

DSA is an important part of creating a safer driving environment. By alerting drivers in real time and encouraging self-correction, it helps prevent risky situations from escalating and reduces the number of violations needing review.

AI Event Types

The GroundCloud VEDR system uses AI to monitor a comprehensive set of driver-facing (DMS) and road-facing (ADAS) safety risks. Each event type listed below represents a specific behavior or hazard the AI is trained to detect in real time. When one of these conditions is observed, the device may issue an audible DSA warning to allow the driver to self-correct. If the behavior continues, or if the situation is too urgent for a warning, the system will report a violation for later review.

Certain DMS event types also include a 20-minute cool-down period, preventing the same event type from being reported more than once within 20 minutes. This ensures drivers are not repeatedly flagged for the same behavior in a short timeframe.

Each DSA warning begins with three audible beeps, followed by the spoken message. These beeps do not precede DSA violation messages; only the violation phrase is heard.

The table below outlines all AI event types currently monitored, along with their descriptions and associated warning/violation messages.

FAQ

What Is the Difference Between MV and AI?

MV stands for Machine Vision – this is the camera side of the system. MV is responsible for seeing what’s happening: it captures images/video and detects visual elements like lane markings, vehicles, pedestrians, the driver’s face, seat belt position, etc.

AI stands for Artificial Intelligence – this is the decision-making side. AI takes the visual input from MV (plus sensor data like speed and motion) and decides what it means:

• Is the driver distracted or drowsy?

• Is the vehicle too close to the one ahead?

• Was there a rolling stop or speeding event?

In short:

• MV = “What do we see?”

• AI = “What does it mean?”

Most of the time, MV/AI is simply referred to as AI.

What Is the Difference Between a Warning and a Violation?

The Driver Safety Assistance (DSA) system uses audible warnings to give drivers a chance to fix unsafe behavior before it becomes a violation.

• If the driver corrects the behavior in time, only a warning occurs and nothing is recorded or reported.

• If the unsafe behavior continues or worsens, the system records a violation, which is uploaded as an event for human review (and may affect KI scoring, where applicable).

How Many Warnings Does a Driver Get?

For certain behaviors (for example, distraction or seat belt), DSA typically provides two warnings before escalating to a violation. For urgent or high-risk situations (such as forward collision or failure to stop), there may be no opportunity for a warning and the device will record the violation immediately.

How Long Does a Driver Have to Correct Their Behavior Before a Violation Is Reported?

It depends on the type of behavior being detected.

• For some behaviors (e.g., no seat belt, using a phone, distracted), the driver typically receives two audible warnings and has a short window to correct their behavior before a violation is recorded.

• For others (e.g., forward collision, stop sign violation), there is no opportunity for a warning and a violation is recorded immediately upon detection of the risk.

Are Warnings Reported to FedEx®?

No. Warnings are not reported to FedEx®.

DSA warnings are played in the cab to give drivers a chance to self-correct, but no record of warnings is stored or sent. Only violations are recorded as events, reviewed and (where applicable) included in reporting to FedEx®.

How Often Will an Individual AI Event Be Reported?

For most AI events, there is no limit to how many times they can be reported during a trip.

However, for three specific event types, Distracted, No Driver and No Seat Belt, FedEx® has defined a 20-minute cooldown period. This means that once a violation for one of these event types is reported, additional violations of the same type will not be reported again until 20 minutes have passed.

How Does VEDR Determine Posted Speed Limits and Vehicle Speed?

For details on where GroundCloud sources speed data, how it’s used and how accuracy issues are handled, see the Speed Data and Speed Events help center article.

My Driver Had Their Seat Belt Fastened. Why Did They Still Receive a Violation?

A seat belt violation can occur even if the buckle is fastened, because the AI detects how the seat belt is worn, not just whether it is latched. For the system to recognize proper seat belt use, the shoulder strap must be securely positioned across the driver’s chest.

If the belt is:

• hanging loosely off the shoulder,

• tucked behind the driver,

• under the arm, or

• not lying flat across the torso

The AI will interpret this as no seat belt detected, and a violation may be reported.

To avoid false violations, ensure the seat belt is worn correctly and snugly across the chest and over the shoulder, as intended by design.

What Is Considered an Incomplete Stop?

An incomplete stop (often called a “rolling stop”) is when a vehicle slows but never comes to a complete stop at a stop sign or required stopping point.

Under FedEx® safety expectations, a proper stop means:

• The vehicle’s wheels come fully to rest

• The vehicle remains stopped long enough for the driver to visually check for cross-traffic, pedestrians, and other hazards

• Only then does the driver proceed safely

If the vehicle continues to creep forward, even at very low speed, without fully stopping, the AI and review team may classify this as an incomplete stop and a violation may be reported.

Is a Scanner Considered an Electronic Device While Driving?

Yes. Under FedEx® safety standards, any handheld electronic device counts as an electronic device while driving, including scanners. This means drivers cannot hold, view or interact with a scanner while the vehicle is in motion.

AI DMS will treat a scanner the same as a mobile phone or similar device. If a driver is seen holding a scanner, looking down at it or interacting with it while driving (even briefly), the system may issue warnings and, if the behavior continues, report a violation.

To avoid violations, scanners must be secured and not handled until the vehicle is fully stopped and parked safely.

I Am a P&D Service Provider. Following Distance Is Not One of My KIs. Why Will My Drivers Still Receive Following Distance Warnings or Violations?

All drivers are subject to the full set of AI safety monitoring features, including Following Distance. Even though Following Distance is not tracked as a P&D KI, P&D drivers may still receive warnings and violations for behaviors such as Close Following or Tailgating. These alerts are part of FedEx’s standard safety expectations for all vehicles and all contract types.

I Am a Linehaul Service Provider. Incomplete Stop Is Not One of My KIs. Why Will My Drivers Still Receive Stop-related Warnings or Violations?

AI safety monitoring applies to all drivers, regardless of contract type. Linehaul drivers may still receive warnings and violations for Stop Sign Violations, including Failure to Stop, Incomplete Stop or Failure to Yield. These alerts are generated because stopping behavior is part of FedEx’s universal safety requirements, even though it is not included in the Linehaul KI structure.

Why Are All Drivers Subject to All AI Event Types, Regardless of Contract Type or Vehicle?

FedEx®applies a single, consistent safety standard across all vehicles and contract types. That means AI safety monitoring (for example, Following Distance, Stop Sign Violations, Seat Belt, Distraction, etc.) is enabled for all drivers, whether they are P&D or Linehaul.

This consistency helps FedEx®:

• Hold all drivers to the same safety expectations,

• Identify and address high-risk behaviors across the entire network, and

• Maintain a clear, defensible safety posture with regulators, insurers, and customers.

Even if a specific behavior is not part of your KI set, your drivers may still receive warnings and violations for that behavior because it is part of FedEx's overall safety requirements.

Are AI Events Reviewed by Humans?

Yes. All events generated by AI or detected via accelerometers are subject to human review.

Do Rental Vehicles Need to Be Equipped with AI-capable Devices?

Yes, per FedEx® guidelines, all equipment that provides service, including rental vehicles, must be equipped with AI-capable VEDR devices.

What If My Drivers Cannot Hear the In-cab Alerts?

The speaker volume on all GroundCloud VEDR devices is set to maximum (Level 10) by default. However, alerts can still be hard to hear in a noisy cab due to engine noise, road noise, wind, radio, or conversation.

To improve audibility, we recommend reducing cabin noise where possible (for example, lowering radio volume and keeping windows closed at higher speeds) so drivers are more likely to hear and respond to alerts.

Are There Factors That May Influence the Accuracy of AI?

Many factors may influence the accuracy of AI, including but not limited to:

• The VEDR device is mounted too far (left or right) from the center of the windshield.

• The camera angles are too high or too low.

• The objects are in view between the subject and driver-facing camera (e.g., sun visor, power cord, radio cord, fan, electronic device mounts, etc.)

• The drivers wear accessories such as hats, eyeglasses, sunglasses and face coverings.

• The Vehicle Type is not set correctly on the vehicle profile.

Does AI Collect Biometric Data?

No. The AI uses camera input to recognize behavior patterns, such as distraction, phone use, or seat belt status. Still, it does not collect, store, or use biometric data (such as facial recognition, identity, or physical characteristics tied to an individual).

Questions or Support

If you still have questions about AI VEDR or need assistance, contact GroundCloud Support at (218) 864-7900 (option 2) or email servicedesk@descartes.com.

Back to Top