How a robot decides where to go
A robot vacuum combines three kinds of information. Range sensing, usually a rotating laser turret or a camera with structured light, builds a map of walls and furniture. Contact and proximity sensing, from a front bumper, wall-following infrared sensors and downward cliff sensors, handles what the map misses. Motion sensing, from wheel encoders and a gyroscope, tracks movement between map updates. Software fuses them, plans a route and stores the result as a map the user sees in an app. An error can enter at any layer, and the user usually reports it as the robot being stupid.
For a brand, navigation is the most visible quality of a robot line and the hardest to sample, because it depends on the home. Two robots with the same suction can score very differently in reviews because one loses its map twice a week.
Causes in order of likelihood
| Rank | Symptom | Typical cause | Layer |
|---|---|---|---|
| 1 | Stops, error voice, cliff or wheel alert | Dirty cliff sensors, wheels lifted by cables, thresholds | Environment and maintenance |
| 2 | Misses rooms or repeats areas | Doors closed, dark floors confusing sensors, mirrors, glass | Sensor type and environment |
| 3 | Map shifts or duplicates | Furniture moved, poor loop closure, wheel slip on rugs | Algorithm and odometry |
| 4 | Cannot return to dock | Dock signal blocked, dock too near wall, low battery | Dock design and routing |
| 5 | App loses connection or map fails to save | Wi-Fi band, router settings, firmware or cloud fault | Software and radio |
| 6 | Collisions, stuck on obstacles | Bumper switch fault, sensor window scratched, laser turret fault | Hardware and QC |
The first two are environment and maintenance. The last four need design and QC attention, and the third and fifth are where firmware quality dominates.
Checks in order
- Wipe the cliff sensors, the bumper and the laser window with a dry cloth and clear hair from the wheels and caster.
- Remove cables, socks and thin rugs from the first test run and open the doors so the robot can see the rooms.
- Place the dock against a wall with clear space on both sides and keep it away from direct sunlight, which can confuse infrared signals.
- Run one mapping pass with the lights on, then compare the saved map with the plan of the home.
- Check the Wi-Fi band the robot supports and update firmware before judging the app.
- If errors repeat with a clean machine and a clear floor, return the unit for sensor and wheel testing.
Sensor choices and what they cost in complaints
| Navigation type | Strengths | Weak points |
|---|---|---|
| Laser distance sensor (lidar) | Works in the dark, accurate room maps | Turret is a moving part, reflective surfaces, height of the turret under furniture |
| Camera with vision | Recognises objects, lower profile | Needs light, privacy concerns, harder to test |
| Structured light or ToF plus bumper | Obstacle detection near the floor | Short range, black surfaces |
| Gyroscope and bump navigation | Simple, low cost | Random coverage, no reliable map |
None of these is universally better. A brand selling to homes with low furniture may prefer a thin camera design, while a brand selling to large open-plan homes will favour lidar. The choice belongs in the specification together with the cliff sensor count, bumper type and wheel suspension travel, because those determine how often the robot meets thresholds and cable tangles.
What it means for importers
Firmware is a deliverable. Map drift, dock return and app stability are software outcomes. The brand needs to know who owns the firmware, how updates are delivered, how bug fixes are scheduled after launch and what the support period is. Ask the partner plant for a defined set of test scenarios, such as thresholds of a stated height, dark carpet, glass doors and multi-room mapping, and have the results recorded on the approval sample.
Wireless rules apply. Robots with Wi-Fi or Bluetooth modules need CE RED for the EU and FCC for the United States, and the module and antenna design affect connection stability. The desk checks that documents match the final firmware and hardware, and verifies them per project. See the certifications guide.
Docks add failure paths. A self-empty dock or mop-wash dock adds contact points, bags, pumps and sensors that can fail. The more the dock does, the more the line needs spare parts and a clear support plan; see warranty and spare parts.
Mechanical wear matters. Cliff sensor windows scratch, bumpers loosen and wheel treads wear, and the effect on navigation is gradual. Spare wheels, side brushes and sensor covers should be orderable.
Test the robot in a real layout
A bench test on a flat floor tells a brand almost nothing about navigation. The useful test is a mock apartment with typical obstacles: a threshold of a stated height, a dark rug, a chair with thin legs, a charging cable, a glass door and a room with the door half closed. The robot is run several times from a clean start, the map is saved and reloaded, and the dock return is repeated from different rooms. The number of stops, wrong turns and failed returns is counted against a limit the brand has set. Because the course is repeatable, the same test can be run on the golden sample and on units from the production lot, and any change in sensors, firmware or wheel suspension shows up as a change in the count. Brands that skip this step usually learn the results from reviews in the first month.
What to specify and inspect
- Navigation method and sensor list, with the number and position of cliff and wall sensors.
- A test scenario list for the approval sample and a pass-fail threshold for the production sample, agreed in writing.
- Firmware version control, update path and a change-notification clause for sensor or module changes.
- Radio documents (CE RED, FCC) verified per project, and the supported Wi-Fi band stated.
- Pre-shipment checks: powered run on a test course, sensor and bumper function, dock return and app pairing on a sample from the lot.
Production variation is real. Sensor windows are glued and aligned by hand in many plants, and a small tilt changes range accuracy. A short calibration step at the end of the line, recorded for each unit, reduces that variation. See testing and inspection and quality control. To scope a robot program, use the quote form.