Four Time-of-Flight cameras each return their own point cloud. Calibration is what turns four overlapping opinions about the same floor into one volume the planner can trust.
This page shows that calibration, the safety fields built on top of it, and the surfaces that still defeat it.
It is tempting to describe an AMR as having "360° vision". It does not. It has a safety layer that must not miss, a 3D layer that measures, and a map layer that remembers. They are built by different sensors because they fail differently.
Answers one question: is something in the way right now? Dual 360° scanners build the protective stop and slowdown zones. This layer is what the safety certification is written against, which is why it does not depend on the map being right.
Answers: what shape is in front of me, and how far away is it? Four ToF cameras return dense point clouds across a volume rather than a single plane — low obstacles, overhangs, and the top of a load all become geometry.
Answers: where am I, and how do I get there? Simultaneous localisation and mapping builds a map on day one and then tracks pose inside it. No tape, no reflectors, no floor markers of any kind.
The recording at the top of this page is the calibration tool, not a marketing render. Four PLY files are loaded at once — one per ToF camera — and each is drawn in its own colour.
In the session shown, the four frames carry between roughly 11,800 and 19,600 points each. Overlay them without alignment and you get four slightly different opinions about where the floor is. Calibration is the process of aligning them into a single Z-up frame, where the floor is the XY plane and Z points up.
The wireframe box is the reference volume the clouds are aligned and cropped into. Everything the planner receives is inside that box.
| Camera | Colour on screen | Points |
|---|---|---|
| tofd4_…_1017 | Cyan | 19,551 |
| tofd4_…_1018 | Pink | 16,177 |
| tofd4_…_8943 | Lime | 14,514 |
| tofd4_…_1019 | Yellow | 11,807 |
A robot carrying a 1,500 kg pallet that overhangs its chassis is, physically, a bigger robot. The protective field has to be bigger too — otherwise the certification describes a machine that is not the one moving.
The platform models and the fork models are equipped differently, because the jobs are different. This is the honest table: what each model actually carries.
| Model | Perception fit | Payload | Built for |
|---|---|---|---|
| S130 | LiDAR + ToF | 60 kg | Light tote and bin transport |
| X13 | Dual LiDAR 360° | 600 kg | Manufacturing line feeding |
| X22 | Dual LiDAR + ToF | 1,000 kg | Pallet and material transport |
| X33 | Dual LiDAR 360° | 1,500 kg | Heavy pallet handling |
| PX33 | 4× LiDAR + ToF | 1,500 kg | Horizontal pallet transport |
| FX33 | 4× LiDAR + photoelectric | 1,500 kg | Stacking and retrieval |
| FX33 Pro | 4× LiDAR + 2× photoelectric | 1,500 kg | US edition stacker |
| CX33 | 3× LiDAR + 2× photoelectric | 1,500 kg | Counterbalance operations |
Every sensor here fails on something. These are the conditions that show up in site surveys, what happens when they do, and what we do about them before go-live rather than after.
| Condition | What happens | What we do |
|---|---|---|
| Glass and mirrors | LiDAR passes through or reflects at the wrong angle; ToF can read a false surface | Map a keep-out zone, or route the lane away from the glazing |
| Black mesh and matte black panels | Returns are absorbed, so the surface is sparse or missing | Physical guard or a mapped no-go boundary |
| Reflective shrink-wrap on loads | Multipath returns put phantom points beside the real pallet | Acceptance test with the actual load, not a demo box |
| Forklift tines and cables | Too thin for a reliable return at distance | Traffic rules and crossing design, not sensor tuning |
| Crowds | Everything works; throughput drops as the robot yields repeatedly | Simulate at expected density, not at empty-floor density |
| Dust, steam, spray | Scattered returns thicken the cloud with noise | Enclosure rating and a wash-down plan for the sensors |
| Direct sunlight | Infrared interference degrades ToF range | ToF is an indoor layer. Outdoor lanes are scoped differently |
| Racking moved overnight | The map no longer matches the floor | Remap after layout changes — it is a 20-minute job, not a rebuild |
None of this is a reason to avoid the technology. It is a reason to survey the floor before quoting a cycle time. Every row above is something a site visit catches and a brochure does not.
We will tell you which lanes the sensors handle cleanly, which ones need a keep-out zone, and which ones no AMR should be driving down.