The other forklift: why vehicle-to-vehicle risk is the blind spot in most safety programs
Most forklift safety programs, and most forklift safety technology, are built around one problem: keeping a vehicle away from a person. In a high-throughput corridor running multiple vehicle classes at once, the more likely collision is between two forklifts — and that's a different physical problem with a different technical answer.
26
Successful mobile-plant prosecutions in Victoria in 2025, second only to falls from height
0.2cm
UWB positioning accuracy in a real factory trial, held 99% of the time
-40%
Drop in GPS accuracy under the same non-line-of-sight conditions UWB held its accuracy
1km
Non-line-of-sight detection range achievable with vehicle-to-vehicle communication, vs ~250m for onboard sensors
Why This Zone Gets Missed
A different problem, wearing the same uniform
Most safety technology was built to solve one specific problem: a person on foot near a forklift. That's the highest-consequence, most visible risk, and it's where regulatory attention and vendor development have focused for years. A corridor running three or four forklifts, an order picker and a ride-on at once tends to get treated as an extension of that same problem — more of the same risk, just with more vehicles in the mix.
It isn't the same problem. Vehicle-to-vehicle collision risk is a distinct physical scenario, and the technology built to detect a person doesn't automatically transfer to detecting another vehicle. Treating it as a variant of pedestrian risk, rather than its own category, is exactly why it gets missed in most site risk assessments.
The Research
Why cameras struggle with this specific problem
A pedestrian-detection camera is built to recognise a person — a fairly consistent shape and movement pattern, seen from in front of a lens with a clear view. A second forklift approaching a blind corner is a harder problem for the same technology, for a reason that has nothing to do with how good the underlying AI model is.
Peer-reviewed research into vehicle perception is direct about this limitation: occlusion cannot be resolved by improving computer vision techniques, because once one vehicle is physically blocked from another's view, it is genuinely unobservable to a camera that cannot see it. No amount of model tuning changes that — it's a property of optics, not an immaturity in the software.
This isn't a hypothetical concern imported from an unrelated field. It's the exact problem the automotive industry spent the last decade addressing through vehicle-to-vehicle (V2V) communication technology, built specifically because onboard sensors, cameras included, are capped by field of view and line of sight — with a practical detection ceiling around 200 to 250 metres, against up to a kilometre for communication-based detection that doesn't depend on either vehicle seeing the other at all.
Camera-based systems are also documented to degrade in low light and adverse weather, and object-tracking research notes that a vehicle which disappears behind an obstruction and reappears can be logged as a new, different vehicle entirely — corrupting exactly the trajectory data a collision-warning system depends on.
The Alternative
Why UWB solves this differently, not just better
UWB-based vehicle detection works on essentially the same underlying principle as V2V communication: each vehicle carries an active device, so detection doesn't depend on either vehicle visually seeing the other. That isn't a marginal improvement on the camera problem — it's a different mechanism that sidesteps occlusion entirely, rather than trying to see through it.
In a controlled factory-floor comparison published in a peer-reviewed engineering journal, UWB positioning held centimetre-level accuracy — within 0.2cm, 99% of the time — even under high metallic interference and non-line-of-sight conditions that degraded GPS accuracy by 40% and LiDAR accuracy by 25% in the same environment.
UWB isn't without its own limitations, and it's worth being honest about them. Published research on UWB positioning consistently flags non-line-of-sight signal reflection — multipath interference — as a source of error, capable of degrading accuracy from centimetre-level to metre-level if left uncorrected. That's precisely why fusing UWB with a second sensing layer, rather than relying on UWB in isolation, is the more resilient design: independent research on sensor fusion for positioning shows combining UWB with a complementary sensing method measurably reduces exactly this class of error compared to either technology alone.
The Pattern
What this means for a multi-forklift corridor
Occlusion is physics, not a software bug
A camera cannot register what it physically cannot see. That's a documented, fundamental limitation of vision-based perception when one vehicle blocks another from view — not a maturity gap that a future software update closes.
The industry that solved this at scale didn't rely on a better camera
The automotive sector's answer to exactly this problem was vehicle-to-vehicle communication technology — a fundamentally different detection mechanism, not an incremental improvement to onboard vision.
Fusion outperforms either technology alone
UWB alone carries its own non-line-of-sight accuracy risk. Pairing it with a second sensing layer is what keeps it reliable in practice, consistent with the wider positioning research on sensor fusion generally.
Where This Leaves the Zone
Same layer, applied to a different pairing
A multi-forklift corridor needs the same underlying answer as any other obstructed, blind zone: detection that doesn't depend on line of sight. The difference is who's being detected. Vehicle-to-vehicle collision avoidance runs on the same sonar and UWB layer already covering pedestrian protection in a Tagged or Hybrid deployment — extended to cover other vehicles fitted with the same devices, not a separate system bolted on to solve this one problem.
A corridor with three forklifts and an order picker in it doesn't need a better camera. It needs a way for each vehicle to know the others are there, whether or not any of them happen to be looking.
This is native to how SonaSafe's Tagged and Hybrid systems work. Vehicle-to-vehicle detection runs on the same sonar and UWB layer used for pedestrian protection, built into the platform from the start — not a separate system added on top to cover this one risk.
Occlusion is physics, not a software bug
A camera cannot register what it physically cannot see. That's a documented, fundamental limitation of vision-based perception when one vehicle blocks another from view — not a maturity gap that a future software update closes.
The industry that solved this at scale didn't rely on a better camera
The automotive sector's answer to exactly this problem was vehicle-to-vehicle communication technology — a fundamentally different detection mechanism, not an incremental improvement to onboard vision.
Fusion outperforms either technology alone
UWB alone carries its own non-line-of-sight accuracy risk. Pairing it with a second sensing layer is what keeps it reliable in practice, consistent with the wider positioning research on sensor fusion generally.
Sources
- Enhancing Autonomous Truck Navigation with Ultra-Wideband Technology in Industrial Environments — peer-reviewed, PMC (National Library of Medicine)
- Vehicle Trajectory Prediction and Collision Warning via Fusion of Multisensors and Wireless Vehicular Communications — peer-reviewed, PMC
- A Systematic Literature Review on Vehicular Collaborative Perception — A Computer Vision Perspective (arXiv, 2025)
- Collaborative Perception for Connected and Autonomous Driving: Challenges, Possible Solutions and Opportunities (arXiv)
- Research on a Visual/Ultra-Wideband Tightly Coupled Fusion Localization Algorithm — peer-reviewed, PMC
- WorkSafe Victoria — 2025 penalties summary
Does your mixed-fleet corridor have this covered, or just the pedestrian side?
Most site risk assessments were written before multi-forklift throughput became the norm. A site assessment checks vehicle-to-vehicle exposure specifically, not just the pedestrian case the traffic management plan was originally built for.
Talk to our team
