AYXWORKS uses necessary cookies to work. Analytics and advertising cookies, including third-party cookies from Google, load only if you allow them. You can change this any time in “Cookie settings” at the bottom of the page.
AGV means Automated Guided Vehicle; AMR means Autonomous Mobile Robot. Conventionally, AGVs follow more prescribed routes while AMRs can plan and adapt paths using environmental sensing. The labels overlap in practice: not every AGV uses floor tape, and an AMR can deliberately operate on constrained routes.
Explanation, examples & sources
Compare required behaviour rather than terminology. Ask how routes are defined, whether obstacles can be bypassed, how traffic is managed and what happens after localisation loss. Both categories need an assessed operating area and suitable protection; “autonomous” does not mean inherently safer. ISO 3691-4 includes both AGV and AMR examples within its driverless-industrial-truck scope. Avoid the common typo “AVG” and vendor claims that one category is always cheaper or superior.
What site preparation does an AMR project require?
An AMR still needs a suitable operating environment. Floors, slopes, thresholds, doors, traffic, load transfer, charging, localisation features and communication coverage can all affect performance. “No fixed guidance infrastructure” is not the same as “no preparation or integration required.”
Explanation, examples & sources
Survey the intended routes during real working conditions, including busy shifts, temporary stock and cleaning. Test the worst floor transitions and docking positions with the actual load. Identify responsibility for doors, lifts, fire routes and shared pedestrian areas. Document map ownership and how layout changes are approved. Start with a bounded route and measurable outcomes before expanding the fleet. A successful empty-warehouse demonstration does not establish reliable operation during production congestion.
How should I measure whether an AMR pilot has succeeded?
Measure completed useful missions under realistic conditions, not just travel speed or distance. Include delivery reliability, waiting time, human interventions, charging, blocked routes and integration failures. The pilot should also establish whether the assessed operating conditions and recovery procedures are practical for the people using the facility.
Explanation, examples & sources
Define the baseline manual process and run the pilot across representative shifts and load types. Record median and slow-tail delivery time, successful transfers and interventions per mission. Investigate whether delays come from the robot, doors, production queues or the fleet manager. Set a clear stop condition for unsafe behaviour and a separate commercial threshold for useful performance. Do not extrapolate a single-robot trial to a large fleet without considering congestion and charging capacity.
Does VDA 5050 make mixed-brand mobile robot fleets plug-and-play?
VDA 5050 defines communication between mobile robots and central fleet control; it does not automatically make every vehicle, docking station or safety design interchangeable. Interoperability depends on the implemented version, supported actions, maps, traffic behaviour and vendor integration. A standards logo is only the start of the compatibility check.
Explanation, examples & sources
Ask each supplier for the exact version and implemented feature set, then test the required missions and exception cases across the chosen fleet manager. The inspected official repository identifies version 3.0.0 and states that the published VDA document governs discrepancies. Do not require that version blindly if installed products support an earlier agreed profile. Include charging, blocked paths, lost communication and order cancellation in acceptance tests. Safety responsibilities remain separate from successful exchange of fleet messages.
LiDAR measures distance using emitted light and its return. A 2D scanner samples a plane; a 3D system samples a volume or multiple directions to build spatial information. The useful choice depends on what must be detected and where it can appear relative to the sensor.
Explanation, examples & sources
A planar scanner can be effective for navigation along walls and fixed structures, but objects above or below its scan plane may be missed. A 3D sensor can represent more geometry while increasing data and processing requirements. Check field of view, mounting, range, update behaviour and surface conditions rather than resolution alone. Sensor data still needs correct timestamps and coordinate transforms. Neither a dense point cloud nor a navigation label establishes a certified protective function.
Why do robot sensors struggle with glass, reflections, dust or changing light?
Sensors depend on physical interactions between the environment and their measurement method. Glass, reflections, dark materials, dust and lighting changes can create missing, misleading or unstable observations. A sensor specification measured under ideal conditions does not guarantee reliable detection of every object in the operating area.
Explanation, examples & sources
Build a challenge set from the actual site: windows, polished metal, low objects, sunlight, hanging material and representative contamination. Test mounting positions and complementary sensing before selecting software thresholds. Record whether a failure causes a missed obstacle, a false stop or poor localisation; each has different consequences. Keep cleaning and inspection in the operating plan. Do not solve repeated failures by disabling detection or reducing protective margins without reassessing the risks and validating an appropriate alternative.
Should a robot use LiDAR, cameras or depth sensors?
Different sensors answer different questions. LiDAR provides range measurements; cameras capture appearance; depth cameras add geometric information through their particular sensing method. A combination can be useful, but only when the added information addresses a real failure mode and the integration is calibrated and tested.
Explanation, examples & sources
Compare the task's needs: reading a label differs from detecting an obstacle or estimating a grasp pose. Test dark, reflective, transparent and low-texture objects under the intended lighting. Consider blind spots, motion blur, interference, processing load and mounting protection. More sensors can also create more failure points and contradictory data. Choose the smallest sensing arrangement that meets the requirement with adequate robustness, and document what it cannot reliably observe.
What do encoders and IMUs contribute to robot navigation?
Encoders measure motion at wheels or joints; an inertial measurement unit (IMU) measures quantities such as angular velocity and acceleration. They help estimate how a robot moves between external observations. Their estimates can drift or be biased, so they are commonly combined with other information rather than treated as perfect position sensors.
Explanation, examples & sources
Wheel slip can make encoder-based travel disagree with actual movement. IMU integration is sensitive to bias and alignment. Check units, axis conventions, mounting and timestamps before tuning an estimator. Use realistic uncertainty rather than presenting every sensor as equally reliable. For debugging, compare raw measurements with independent observations on simple straight and turning motions. A map cannot compensate for badly configured odometry if the entire navigation stack receives inconsistent motion information.
What is the difference between mapping, localisation, SLAM and path planning?
Mapping builds a representation of the environment. Localisation estimates the robot's position within a reference or map. Simultaneous Localisation and Mapping (SLAM) estimates both together. Planning chooses how to reach a goal, while control executes the movement. These functions cooperate but are not the same algorithm.
Explanation, examples & sources
A robot can localise in a previously created map without continuously rebuilding it. It can also have a good position estimate yet fail to plan through a narrow aisle, or plan a valid route but track it poorly. Debug by identifying the failing layer rather than calling every navigation issue “bad SLAM.” Preserve maps and configuration with version information so changed behaviour can be traced to a deliberate update rather than an unexplained field adjustment.
Can SLAM work in a changing warehouse or around moving people?
It can, within the capabilities and validated operating conditions of the system. Dynamic objects do not necessarily prevent localisation, but large layout changes, obscured landmarks and repetitive geometry can reduce reliability. A system that maps an empty building successfully may behave differently during a busy shift.
Explanation, examples & sources
Distinguish stable structural features from temporary stock and moving traffic. Test realistic congestion, rearranged pallets, doors and seasonal conditions. Define who may update the map and how routes and protection are rechecked afterward. Avoid continuously accepting every observed change without understanding its effect on localisation and navigation. Record the conditions under which the robot must stop and request assistance. Operational discipline around layout changes can matter as much as choosing a more advanced algorithm.
How do LiDAR SLAM, visual SLAM and sensor fusion differ?
LiDAR SLAM primarily uses range geometry; visual SLAM primarily uses information from camera images. Sensor fusion combines complementary measurements, such as wheel motion, IMU and external observations, to improve an estimate. The strongest approach depends on the environment, motion and failure cases, not a universal ranking of sensor types.
Explanation, examples & sources
Feature-poor corridors, repeated visual patterns, reflective surfaces or changing lighting can challenge different methods. Evaluate the intended route and task rather than an unrelated benchmark dataset. Good fusion also requires calibrated frames, time alignment and sensible uncertainty. Combining two biased or misconfigured sources can make an estimate worse. Prefer a diagnosable configuration with recorded test cases over a more complex stack whose failures cannot be traced.
Why can a robot know its map position but still miss a docking or picking target?
Global navigation accuracy is not necessarily sufficient for a precise local task. Wheel slip, map uncertainty, mounting tolerances and coordinate errors can accumulate. Docking, charging or robot picking may therefore need mechanical guidance, local markers or close-range sensing in addition to the global pose estimate.
Explanation, examples & sources
Specify the tolerance at the actual interface, including orientation and load, instead of requesting an impressive map-position number. Measure repeatability across approaches, floor conditions and battery states. Check the transform chain from map to base, sensor, tool and target. A stable-looking display can hide an incorrect frame or timestamp. Separate the acceptance criteria for reaching the workstation from those for making a precise transfer, and give each a recovery procedure.