A robot designed for a climate-controlled office in a temperate country meets rather different conditions in a Vietnamese restaurant. Six local factors that determine whether a machine performs as specified, and what to ask about each.
Heat and humidity
Heat reduces performance before it causes failure. Processors slow themselves to stay within limits, and a robot that responded quickly in a demonstration room may respond noticeably slower on a hot afternoon in a busy space.
Batteries lose capacity in heat and degrade faster over their life. A stated runtime measured at moderate temperature will not be achieved in a hot environment.
Humidity affects electronics over time, particularly where a machine cycles between air conditioning and outdoor humidity, because condensation forms.
What to ask. The rated operating temperature range, and whether the runtime figure was measured at the top of it. Whether the manufacturer has deployments in tropical conditions.
What to do. Keep charging areas ventilated and out of direct sun. Do not park a robot against a hot wall. Where a machine moves between conditioned and unconditioned areas, expect a shorter life than the specification implies and plan for it.
Power and network
Power interruptions are a normal condition, not an exception. A robot mid-task when power fails should resume sensibly. Ask what happens — some models lose their position and need re-mapping.
Voltage variation matters for charging equipment. Where supply is unstable, protection on the charging circuit is cheap insurance.
Most conversational robots depend on the network. The language processing usually runs on a server, so when connectivity fails the robot becomes markedly less capable — sometimes mute.
Ask what it does offline. A robot that can still navigate, still deliver, and still show preset information on its screen degrades gracefully. One that stops entirely does not.
Wireless coverage across the whole route. A robot that works near the router and stalls at the far end of the building is a coverage problem being misdiagnosed as a robot problem. Test the full path before purchase, not after.
Floors and physical layout
Floor surface determines whether it moves reliably. Polished tile is good. Loose tiles, thresholds between rooms, and door tracks are the common obstacles. A threshold of even a couple of centimetres stops many service robots.
Wet floors are common in restaurants and shops, and they affect traction and can damage a machine not rated for them.
Slopes and ramps. Check the rated gradient against the actual ramps on site, including the one at the entrance.
Narrow passages. Robots need more clearance than their width suggests because they turn. Measure the tightest point of the intended route.
Lifts. If the route includes one, the robot needs a way to call and use it, which usually means integration work and sometimes building modifications.
Walk the route with a tape measure before buying. An hour spent doing so has prevented a great many unusable purchases.
Crowds and noise
Vietnamese public spaces are busier and noisier than the environments most service robots were designed and tested for. This affects two things directly.
Navigation. A robot in a dense crowd may stop repeatedly and make little progress. Some handle this well and some effectively freeze. Ask specifically how it behaves in crowds, and observe it in a busy setting rather than an empty demonstration space.
Speech recognition. Background noise is the single largest cause of misrecognition. A robot that understands perfectly in a quiet room may struggle at peak hours in a restaurant.
What helps. A directional microphone array. A screen offering preset options so customers are not dependent on speech. Placement slightly away from the noisiest area.
People interact more physically here. Children touch, adults photograph, people step in front of it deliberately to see what it does. A machine with exposed sensors or delicate external parts will not last.
What to specify differently
Drawing the practical conclusions together.
Ask for the runtime figure at 32 degrees, not at 20. Or apply a discount to the quoted figure and plan accordingly.
Require a defined offline behaviour and test it by disconnecting the network during evaluation.
Check the threshold and gradient ratings against the actual site, not against a general description of it.
Observe the model in a genuinely busy environment before committing. Crowd behaviour cannot be assessed any other way.
Favour robust external construction over refined appearance, given how physically people interact with them here.
Confirm parts are held locally. A component ordered from abroad means weeks out of service, and weeks out of service in months three to five is when a deployment dies.
None of this is exotic. It is simply testing against local conditions rather than against the specification sheet, and it is the difference between a machine that works here and one that worked where it was designed.
Frequently asked questions
How does heat affect a robot's stated performance?
Processors slow themselves to stay within limits, so responses take longer, and batteries lose capacity and degrade faster. A runtime figure measured at moderate temperature will not be achieved in a hot environment.
What happens to a conversational robot when the network fails?
Most become markedly less capable, sometimes mute, because language processing runs on a server. Ask specifically what it can still do offline and test by disconnecting during evaluation.
What physical obstacles stop service robots most often?
Thresholds between rooms and door tracks — even a couple of centimetres stops many models. Walk the intended route with a tape measure before buying, including the tightest turning point.
Why do crowds matter more here?
Vietnamese public spaces are busier and noisier than the environments most service robots were tested in, which affects both navigation and speech recognition. Observe the model in a genuinely busy setting before committing.
More in Myths and questions and Robots and jobs.
