SHARE:
When companies approach me about AGV or AMR projects, the first thing I usually see is confusion—not about robots, but about sensors. Most teams understand that sensors matter, but they underestimate how deeply sensor selection affects navigation reliability, safety compliance, uptime, and long-term operating cost. In my experience, successful projects don't start with brands or specifications; they start with a clear framework that separates engineering reality from safety responsibility and procurement risk.
In this article, I walk through how I personally approach AGV and AMR sensor selection in real manufacturing environments. I focus on what actually fails on factory floors, how standards shape design decisions, and why a proper sensor stack is about balance, not maxing out specifications.
Before I ever discuss sensors, I make sure everyone in the room agrees on one thing: AGVs and AMRs are not just different marketing terms—they represent fundamentally different autonomy models. That distinction alone dictates how dependent the vehicle will be on its sensors.
Traditional AGVs follow fixed paths using magnetic tape, QR codes, or reflectors. In these systems, sensors play a supporting role. They help with obstacle detection, docking confirmation, and safety, but they are not responsible for real-time decision-making. Because of that, AGV sensor stacks tend to be simpler, more predictable, and easier to validate.
AMRs, on the other hand, must perceive and interpret their environment continuously. Localization, obstacle classification, dynamic path planning, and human interaction all depend on sensor input. If an AMR loses sensor confidence, it doesn't just slow down—it may lose its position entirely or make unsafe decisions.
This is why I always say: the more autonomy you demand, the more your project becomes a sensor project. Teams that treat AMRs like “AGVs without tape” usually discover this the hard way—during commissioning or, worse, during production ramp-up.
One of the most dangerous misconceptions I encounter is the idea that “good navigation sensors are enough for safety”. From an engineering and compliance perspective, this is simply wrong.
Navigation sensors exist to help the robot understand where it is and where it can go. Safety sensors exist to protect people. These two responsibilities are governed by completely different rules, failure tolerances, and certification requirements. No matter how advanced a LiDAR or vision system is, it cannot replace a safety-rated device unless it is explicitly certified for that purpose.
From a system design standpoint, I always separate these sensor systems at both the hardware and functional level. Navigation sensors feed into the autonomy stack. Safety sensors feed into a safety controller that can stop the vehicle regardless of software state.
This separation matters not only for compliance, but also for fault handling. When a navigation sensor degrades, the robot may slow down or re-localize. When a safety sensor detects a fault, motion must stop—immediately and predictably. Mixing these responsibilities creates ambiguity, and ambiguity is the enemy of functional safety.
When people ask me which sensors are “mandatory”, my answer is always the same: it depends on environment, risk, and performance targets. That said, there are a few sensor categories that appear in almost every serious AGV or AMR deployment.
LiDAR remains the backbone of most AMR navigation systems because it performs reliably across lighting conditions and provides accurate geometric data. Cameras add semantic understanding—recognizing pallets, humans, and markers—but they struggle in poor lighting or dusty environments. Ultrasonic and proximity sensors act as short-range redundancy, particularly for low-speed maneuvers and docking. IMUs and wheel encoders, while often overlooked, are critical for maintaining localization stability when external sensing is degraded.
What I emphasize to teams is that every sensor has a failure mode. LiDAR struggles with glass and reflective surfaces. Cameras struggle with glare and shadows. Encoders drift over time. A good sensor stack is not about choosing the “best” sensor—it's about ensuring that no single failure mode brings the system down.
Most failures don't come from bad technology; they come from unrealistic assumptions. I've seen projects designed on clean CAD layouts that completely ignore lighting variation, airborne dust, floor reflectivity, or human behavior. Sensors that perform perfectly in a lab often behave very differently in a live factory.
One common mistake is selecting sensors based solely on range or resolution, without considering stopping distance at full speed. Another is ignoring contamination—oil mist, packaging dust, or condensation that gradually degrades sensor performance. I've also seen teams underestimate maintenance access, placing sensors in locations that are difficult to clean or recalibrate.
In my experience, the biggest red flag is when sensor selection happens before site conditions are fully understood. The environment should always define sensor thresholds, not the other way around.
KH Group Fork-type AMR - KHFP150 & KHFP200 Series
No serious deployment relies on a single sensor type. Sensor fusion is not optional in real factories—it is mandatory. Over time, I've seen certain patterns emerge across industries.
In warehouses, 2D LiDAR combined with cameras and safety laser scanners is common. In manufacturing plants with heavy machinery and narrow aisles, redundant LiDAR layers and robust safety scanners dominate. Outdoor or mixed-environment AMRs often add 3D LiDAR to handle elevation changes and irregular obstacles.
The key is that each sensor layer serves a distinct role: global localization, local obstacle detection, short-range confirmation, and safety enforcement. When these roles are clearly defined, system behavior becomes predictable—and predictability is what operations teams value most.
From a procurement and compliance standpoint, safety standards are not optional reading—they are design inputs. ISO 3691-4, in particular, defines how driverless industrial trucks must detect humans, manage stopping distances, and handle faults.
What I often explain to buyers is that compliance is not just about passing an audit. It's about limiting liability and ensuring that the system behaves safely even when things go wrong. This is why safety-rated laser scanners, redundant circuits, and validated safety functions add cost—but also reduce long-term risk.
Procurement teams should be asking suppliers not just what sensors they use, but how those sensors are validated, tested, and maintained over time. Acceptance testing, safety documentation, and change-management procedures are just as important as hardware specifications.
KH GROUP Backpack handling robot—KHD80
After reviewing dozens of projects, I see the same mistakes repeated again and again. The most common is single-sensor dependency—assuming one technology can handle all scenarios. The second is confusing detection with safety compliance, especially when vision systems are involved.
Another frequent error is ignoring total cost of ownership. A cheaper sensor that requires frequent cleaning, recalibration, or replacement often costs more over the system's lifetime than a higher-quality alternative. Finally, many teams underestimate the organizational impact of sensors, forgetting that maintenance staff must understand and support these systems long after installation.
Avoiding these mistakes doesn't require cutting-edge technology. It requires disciplined engineering, honest risk assessment, and close collaboration between engineering, safety, and procurement.
When I look at AGV and AMR sensor selection, I don't start with products—I start with responsibility. Navigation, safety, operations, and procurement each have different priorities, and a successful sensor stack respects all of them. By separating sensor roles, designing for real-world failure modes, and aligning with safety standards from day one, companies can avoid costly redesigns and delayed deployments.
If you're planning an AGV or AMR project and want a sensor strategy that holds up beyond the demo phase, I always recommend stepping back and evaluating the full system—not just the robot. The right sensor decisions made early can save years of operational headaches later.
Copyright © 2025 KH AUTOMATION PTE. LTD. All Rights Reserved KH GROUP
