Unsafe riding, caught on the bike.
RideGuard is an AI unit for delivery motorbikes. It detects 19 kinds of violations and unsafe riding on the unit itself, and turns each one into a signed record with the clip, location and telemetry needed to act on it.
- On the unit19 detectors and four sensors fused 30 times a second. No server needed for detection.
- For authorities and fleetsA signed record of every violation: clip, location and telemetry, ready to act on.
- For the riderOptional: a spoken warning through the app and headset, within a second.
Simulated camera view with the detections the unit draws in. Riders using the app also hear each warning through their headset.
Signed violation log
Every record is hashed and linked to the one before it, ready to act on.Why monitoring has to happen on the bike
Accidents involving delivery riders keep rising, and Dubai is moving to cameras and telematics on delivery bikes.
70,166 rider violations were recorded in 2024.
In October 2025 RTA announced cameras and telematics on 12,000 delivery bikes, and said platforms will have to adopt such a system once trials end.
Fast lanes are now closed to delivery bikes, and speed is capped at 100 km/h.
Accidents involving delivery riders in Dubai
Before the end of October 2025, the year’s count had already passed the whole of 2024. Figures as reported by Dubai Police.
Evidence
When a rider disputes a fine or a crash, the fleet has nothing but the rider’s word.
Behaviour
Warnings that arrive after the shift change nothing. The rider needs to hear it while riding.
Four sensors, one picture of the ride
Every reading is fused 30 times a second, checked against the others and handed to 19 detectors. Everything runs on the unit, so detection never waits on a server.
Wide camera and AI chip
Objects and rider pose, 30 fps
GPS
Position, speed, heading, time
LiDAR
Distance to the vehicle ahead
Motion sensor
Acceleration, lean, impacts
Map data
Limits, one-way streets, lanes, zones
Fusion, 30 times a second
- One combined picture of the ride
- The last ten seconds kept in memory
- Sensors cross-check each other
19 detectors, 10 times a second
Speed, direction, zones, hours, distance, overtaking, posture, passenger, backpack and crash
Signed violation record
7 s clip, telemetry, signature
Dashboard
Map, health, violation feed, risk score
Rider warning, optional
Through the app and headset
The fleet connects to the unit over a private network to read it.
One unit, mounted on the delivery box
It runs from the bike’s 12 V supply and starts by itself when power comes on. Select a marker to see what each part does.
Select a marker. The blue box on top of the delivery box is the unit; the shaded wedge is the camera’s view and the yellow line is the LiDAR beam.
| Edge computer | Runs everything on the unit and starts by itself at power-up |
|---|---|
| AI accelerator, 40 TOPS | Object detection and rider pose, up to 30 frames a second |
| Wide-angle camera, 120° | Sees its own rider and the road ahead |
| GPS receiver | Position, speed, heading and time, once a second |
| Lane-level accuracy | Designed to meet the sub-metre benchmark for lane violations |
| LiDAR rangefinder | Distance to the vehicle ahead, up to 12 m |
| Motion sensor, ±32 g | Acceleration, lean angle and impact peaks |
| Power | From the bike’s 12 V supply |
Measured on the unit
- Vision
- 30 fps on an open road, 23 fps with rider pose running
- Fusion and detectors
- 30 Hz and 10 Hz, with ten seconds of history
- Load
- About 430 MB of memory and a quarter of one CPU core
- Dashboard
- Responds in under 5 ms on the local network
- Maps
- Road limits, one-way streets, lane counts and no-riding zones from OpenStreetMap, built per city with one command
Built for the 18 use cases, plus crash detection
Every detection becomes a violation record with its evidence attached. Each detector names the sensors it relies on and how long the evidence must hold before it fires. Select a row for the detail.
Hold time is how long the evidence must stay above threshold. Camera and zone detectors also take a second or two to build their score first.
It only fires when the sensors agree
A single reading never triggers an alert. Pick a situation that fools simpler systems and see which check catches it.
Sustain timers
Nothing fires on one reading. Speed and direction need 3 s, hours and off-network 10 s, parking 60 s.
GPS cross-check
A speed reading is rejected if the motion sensor says the bike is still, or the last 12 s of track don’t match.
Map sanity
No road within 60 m means no limit is enforced. More than 5 km from any road means the wrong map, so nothing fires.
Divided roads
Wrong-way fires only when every nearby road is one-way against the rider.
Bike in frame
One-handed, passenger and backpack need a two-wheeler under the person, held across frames.
Riding required
A rider stopped on a footway is not riding on it.
Crash needs riding
Picking the unit up off a desk is not a crash.
Cooldowns
Each code fires at most once per 15 s to 10 min, so one situation produces one event.
Riders can hear it while riding
Optional for each fleet. Riders sign in to the RideGuard app, which shows speed against the road’s limit and speaks a short warning through their headset within a second. Try a situation.
- One large speed number, coloured green, yellow or red against the limit of the road the rider is on, with the limit shown as a road sign.
- Spoken phrases for 13 codes. “Crash detected” and “Wrong way” interrupt everything else; the rest queue behind them.
- Warnings play through the rider’s headset, so eyes stay on the road.
- Each rider signs in with their own account, so every warning and violation is tied to the right person.
Mock-up with sample data.
Every violation, ride and sensor in one view
A live map, sensor health, a 0 to 100 risk score per ride, and a violation feed where every entry plays its evidence clip.
System health
- GPS
- 11 sat, hdop 0.9
- LiDAR
- 940 cm
- Vision
- 23 fps, 3 obj, 1 pose
- Motion
- 1.02 g, lean 8°
- Fusion
- 29.9 Hz
- Evidence
- 212 clips
Ride risk
- Speeding
- 34.5
- Hard acceleration
- 2.8
- Lean
- 0.7
- Violations
- 4.0
Violation feed
Mock-up with sample data on the Dubai road network. From the settings drawer the fleet can switch detectors on and off live, fire a test event, change the configuration (for example the road-network file) and restart the unit remotely.
Evidence that holds up when a fine is disputed
Every violation is saved with a clip, a telemetry timeline and a signature, and linked to the record before it. Edit or delete one record and the chain shows exactly where.
Clip
About seven seconds of video around the event, starting before it began, with boxes, labels and rider skeletons drawn in. Encoded on the unit within about three seconds.
Timeline
Twenty points of GPS position, speed, heading, signal quality, acceleration and lean, attached to the event.
Signature
Each event is hashed, signed with the unit’s own key and linked to the previous one. The clip’s hash is stored with it.
Try to tamper with the record
This demo computes real SHA-256 hashes in your browser. On the unit, each record is also signed with the unit’s private key.
One-request verification
A single request re-checks every signature, link and clip on the unit. The current store verifies clean: 1,000 signed events, 0 breaks, in 0.3 s.
Independent check
The unit’s public key is published, so a third party, such as an authority reviewing a case, can verify an event on its own without access to the unit.
Retention
The newest 600 clips or 2 GB stay on the unit; older ones are pruned.
Tested scenario by scenario
67 scenarios played through the production software with real Dubai map data and the sensor signals each ride produces, plus system tests on the hardware, including a 16-day soak.
- 0 msDetectionA detector’s hold time is met
- 17 msRider warnedAlert reaches the phone
- 3.0 sClip encodedBoxes, labels and skeletons drawn in
- ThenStored and signedHashed, signed and linked
Scenario tests
A selection of the 67: what should happen, and what the unit did.
System tests
- Unit and regression suite
- 120 tests pass in 30 s, run on every change
- End to end
- Alert on the phone in 17 ms; clip encoded in 3.0 s; event stored, hashed and signed
- Chain-of-custody check
- 1,000 signed events, all signatures valid, 0 broken links, in 0.3 s
- Remote detector switch
- Off and back on in 10 ms, setting kept across restarts
- City switch
- One file edit, back online in 25 s with 24,910 road segments indexed
- Power loss
- Service starts on its own at every power-up, with a new trace file and a start event each time
- Sensor unplugged
- Alarm within 6 s; detectors go quiet instead of wrong; recovery logged
16-day soak
45,000 health samples. Sensor fusion held at 29.9 Hz on average and never dropped below 28.8 Hz, with zero detector errors.
Data stays on the unit
Detection, clips and the evidence store all live on the unit. The fleet reads them over a private network, and nothing is needed from a server for detection.
- Clips are kept on the unit: the newest 600 or 2 GB, with older ones pruned.
- Anyone with the unit’s published public key can confirm an event hasn’t been altered.
- Detectors can be switched on and off remotely, and the setting survives restarts.
- Retention and access will follow the data protocols agreed in RTA’s ITD review.
Reads the live map, sensor health, violation feed and clips.
Verifies any event with the published public key, without access to the unit.
On the bike and off it
RideGuard covers what happens on the road. Boolanga’s compliance platform already covers what happens before the shift, for delivery fleets across the region.
Before the shift
Boolanga compliance platform, in use today
- Daily rider inspections: appearance, uniform and hygiene
- Gear condition: helmets, bags, pads, reflectors and jackets
- ID and documentation checks
- Violation history and scoring across riders, fleets and 3PLs
On the road
RideGuard
- 19 detectors running on the bike
- Spoken warnings to the rider within a second
- Signed video evidence for every event
- A 0 to 100 risk score for every ride
One view of each rider
Gear, identity and riding behaviour together, so a fleet or regulator sees the whole picture.
Path to deployment in Dubai
Every vendor goes through the same approvals. This is where RideGuard stands.
Unit built and tested Done
Hardware and software complete, with 67 scenario tests and a 16-day soak on the unit.
Registered on the RTA supplier portal Done
Set up for the approval and proof-of-concept stages.
Technical presentation and evaluation unit Next
A walk-through for the review team, then a unit shipped for RTA’s own testing.
SIRA device approval
The device is submitted for approval.
RTA ITD review
Security, platform compatibility and data protocols.
On-road proof of concept
Monitored by RTA under real operating conditions in the UAE, including environmental stress.
Rollout to fleets
Open to operators and 3PLs across Dubai’s 70,000+ delivery bikes.
See it on a bike
We can walk your team through the unit and its test results, or ship one for your own testing under real operating conditions.