If a home robot dashboard has twenty-five numbers on it, the operator usually watches three.

That is not because the other data is useless. It is because most metrics describe activity while only a few change a decision. “Tasks scheduled,” “maps created,” “distance traveled” and “hours powered on” can explain what the robot did. They do not tell you whether the household should keep using it, change the route, move the dock, shorten the task, call support or stop a risky behavior.

For day-to-day operation, the most useful metrics answer four questions:

  1. Did the robot finish the job that mattered?
  2. How often did a person have to rescue it?
  3. When something failed, how quickly did the system recover?
  4. How much maintenance and supervision did the result actually cost?

Everything else should earn its place by helping answer one of those questions.

IEC 62849:2025 is useful context because it provides performance testing and evaluation methods for certain indoor household and similar-use robots. It explicitly says it is not a safety standard and does not set performance requirements. ISO 13482:2014, by contrast, addresses safety requirements for specified personal-care robot categories. Those boundaries matter: an operator dashboard can help you observe performance, but a good dashboard is not a certification and cannot substitute for product-specific safety instructions or applicable standards.

Start with task success, not uptime

The first metric should be brutally simple:

Task success rate = completed intended tasks ÷ attempted intended tasks

The word “intended” matters.

If a robot is supposed to carry a small item from the kitchen to a bedroom, “it drove for 12 minutes” is not success. If it reached the wrong doorway and waited there, the activity log may look busy while the task failed.

Define success before collecting data. For example:

  • delivery task: item reaches the correct destination without being dropped;
  • patrol task: required checkpoints are visited and the robot returns to the dock;
  • reminder task: the reminder is delivered at the right place and time;
  • fetch/assist task: the requested object or interaction is completed within the allowed operating boundary.

Do not combine all tasks into one average if they have different difficulty or risk. A robot can show a 92% overall success rate because easy patrols dominate the denominator while a high-value assistance task fails half the time.

Use a small table by task class.

Task class Attempts Successful Human-assisted Failed Decision
Routine route 40 38 1 1 Keep
Object delivery 20 14 4 2 Investigate route/handling
Timed reminder 15 15 0 0 Keep
New experimental task 6 2 3 1 Pause expansion

The table makes the weak task visible instead of hiding it in an overall average.

Measure intervention rate because autonomy can be fake

A robot can technically “complete” a task only because somebody nudged it around a chair, reopened a door, cleared a cable, reissued a command or picked it up and moved it.

That is not the same operating value as autonomous completion.

Track:

Human intervention rate = tasks requiring human rescue ÷ attempted tasks

Then classify the intervention. A useful taxonomy is:

  • environment: door, rug, cable, clutter, lighting, threshold;
  • navigation: localization, route planning, obstacle handling;
  • manipulation: grip, pickup, handoff;
  • software: app, network, account, update, command failure;
  • human-process: unclear setup, wrong schedule, household member moved an item;
  • safety stop or operator precaution.

The category matters more than the total. If most interventions are caused by one loose rug, the fix is cheap. If they are scattered across localization, software and docking, you may have a product or environment-fit problem.

A falling intervention rate over the first few weeks is a good sign: the home and the robot are learning to coexist. A rising rate after a software update or environment change deserves attention.

Track recovery, not just failure

Failures are inevitable in a real home. The useful question is what happens next.

Measure two recovery numbers:

Autonomous recovery rate — how often the robot resumes correctly without a person.

Median time to recover — how long the household waits before normal operation returns.

This exposes a difference that a simple “error count” misses. Ten small errors that clear themselves in five seconds may be less disruptive than one error that requires a family member to reset the device, reconnect Wi‑Fi and remap a room.

For severe or safety-relevant events, do not let a median hide the tail. Record each event separately with timestamp, task, environment condition and operator action.

Docking deserves its own metric

Many household robots depend on a dock or charging location. A robot that works well during the task but repeatedly fails to return or recharge creates a supervision tax.

Track:

  • successful dock returns;
  • failed approaches;
  • manual placement on the dock;
  • charging interruptions;
  • “not ready” starts caused by incomplete charging.

A docking problem can look like a battery problem. If the robot starts the day at low charge because it missed the dock overnight, replacing the battery does not fix the root cause.

Before changing hardware, inspect the dock location, nearby obstacles, floor surface, alignment and any product-specific clearance instructions.

Maintenance minutes reveal the real labor trade

Households do not buy a robot because they want another machine to manage.

Track weekly maintenance minutes separately from active human rescue. Include cleaning sensors, removing hair or debris, checking wheels, emptying or filling consumables, installing approved parts, applying updates and handling routine calibration required by the manufacturer.

Then compare maintenance with the useful labor displaced.

A robot that saves 90 minutes of work but requires 15 minutes of maintenance may be valuable. A device that saves 30 minutes and consumes 35 minutes in cleaning, remapping and recovery is automation only on the brochure.

Do not turn this into a universal ROI claim. Different household tasks have different values, and some robots are bought for independence, consistency or safety-related assistance rather than labor savings alone.

Measure blocked-space rate when the home is the bottleneck

A home is not a warehouse. Chairs move. Bags appear. Doors close. Pets sleep in routes. Lighting changes. People leave charging cables in exactly the wrong place.

Instead of blaming every failure on “AI,” measure how much of the planned operating area is regularly unavailable.

A practical blocked-space metric can be:

Blocked required locations ÷ required locations observed during the run

You can also track the top five recurring environmental blockers by count.

This metric turns vague complaints—“it gets stuck a lot”—into a household action list. Maybe the robot does not need smarter navigation; maybe the route needs one furniture change and a cable channel.

Add a quality metric for the actual task

Generic robot metrics are not enough. Every task needs one domain-specific quality measure.

Examples:

  • delivery: correct item and correct destination;
  • patrol: required checkpoints covered;
  • social/reminder use: correct person/time/context;
  • mobility or manipulation assistance: completion within the approved operating scenario and any required supervision;
  • cleaning robots: use the product-specific or applicable cleaning metrics rather than pretending this article’s generic framework replaces them.

IEC 62849:2025 has a defined scope for certain indoor floor robots and specifically excludes wet/dry surface-cleaning robots from that document. That is a good reminder to respect the scope of a standard instead of attaching one “robot score” to every machine.

Safety events are a separate register, not a KPI to optimize downward

Do not compress safety-relevant observations into a pretty rate.

Maintain a separate event log for:

  • collisions with a person or pet;
  • unexpected motion;
  • unsafe reach or manipulation;
  • repeated near-miss at stairs or edges;
  • overheating, charging or battery anomalies;
  • behavior outside the intended operating area;
  • emergency stop or manual stop used for precaution.

Follow the manufacturer’s instructions and applicable safety guidance. ISO 13482:2014 covers safety requirements for specified personal-care robots such as mobile servant robots, physical assistant robots and person carriers, while excluding categories including medical devices and industrial robots. A consumer dashboard cannot determine conformity.

The metric is for escalation and traceability, not for proving safety.

A weekly dashboard that is small enough to use

A useful weekly review can fit on one screen:

Outcome

  • success rate by task class;
  • quality metric for each important task.

Autonomy

  • human intervention rate;
  • top intervention cause;
  • autonomous recovery rate;
  • median recovery time.

Readiness

  • dock success rate;
  • failed starts due to charge/setup;
  • weekly maintenance minutes.

Environment

  • recurring blocked locations;
  • changes in the home that affected operation.

Exceptions

  • safety-relevant event log;
  • software/update changes;
  • unresolved support cases.

Then add a decision next to every red metric: keep, adjust, investigate, pause or escalate.

That last column is what makes the dashboard operational. A metric that never changes a decision is probably a diagnostic field, not a headline KPI.

Put market context in the right place

The International Federation of Robotics separates consumer service robots into groups that include domestic tasks, social interaction and education, and care at home. Its World Robotics 2026 materials also show that the service-robot market is broad and growing, but market growth does not tell you whether a specific robot works in a specific home.

Use market data to understand categories and supplier activity. Use household operating data to decide whether your deployment is succeeding.

Those are different questions.

The bottom line

A home robot is useful when it completes valuable tasks with tolerable supervision, recovers predictably, stays ready for work and does not consume more household attention than it saves.

So start with six numbers:

  1. task success by task class;
  2. human intervention rate;
  3. autonomous recovery rate;
  4. recovery time;
  5. dock/readiness success;
  6. weekly maintenance minutes.

Then keep task quality and safety-relevant events beside them, without pretending either can be reduced to one universal score.

If the dashboard cannot tell you what to change on Monday morning, it is probably measuring the robot for curiosity rather than operating it for value.

Sources

Related Reading