What we measure

The fleet already publishes enough to answer useful questions about how it behaves. Roboblackbox turns that runtime record into measurements that operators can inspect over time.

Reported, never judged.

Available measurements depend on the streams the fleet publishes, the capture scope, and the quality of the record.

1 · People and proximity

How close do robots get to people?

Closest approaches

What we observe: robot position and person position as published by the fleet.
What we derive: minimum separation, distribution over time, and event frequency.
What we do not claim: whether the published person position exactly matched physical reality.

Speed while people are present

What we observe: published robot speed and person-presence signals in the same time window.
What we derive: speed distributions and time spent above a configured limit while people are reported present.
What we do not claim: that every person was detected, or that a measured speed was appropriate for the situation.

Findings card for the closest approaches of the month, with the closest, the rule, and the count by distance band — Product preview · synthetic test data

Closest approaches by place, with the count under the rule, the closest distance, the robot, and the July count — Product preview · synthetic test data

2 · Zones and operating behavior

Findings card for too-fast events near people, by aisle and shift, with the worst event — Product preview · synthetic test data

Too-fast events by place and shift with the rate per 100 hours, the worst speed against the rule, the robots accounting for most, and evidence references — Product preview · synthetic test data

Zone exceedances

What we observe: published robot position and speed, alongside the declared zone boundaries and limits.
What we derive: where, how often, and for how long recorded motion exceeded the configured zone limit.
What we do not claim: that the configured boundary or limit matched conditions on the floor.

Exposure by zone and operating context

What we observe: published motion, location, and available shift or person-presence context.
What we derive: hours of motion and event rates by zone or operating context, with the denominator stated.
What we do not claim: that different zones or shifts are directly comparable without accounting for their operating context.

3 · Control integrity

Do robots do what they were told?

Commanded vs actual motion

What we observe: published motion commands and reported motion, aligned within the available clock accuracy.
What we derive: the size, duration, and frequency of differences between commanded and reported motion.
What we do not claim: why a difference occurred, or that reported motion is an independent physical measurement.

Stop-response latency

What we observe: a published stop command and the subsequent reported motion.
What we derive: elapsed time until reported motion meets the stated stop threshold, with timing limits disclosed.
What we do not claim: a certified braking time, or that a command reached the controller when it was recorded.

Findings card for commanded against reported motion, with the share that matched and the nine mismatches by kind — Product preview · synthetic test data

4 · Safety systems

Where do safety stops concentrate?

Protective-stop frequency

What we observe: protective-stop state transitions published by the fleet.
What we derive: stop counts and rates over recorded operating time, using a stated event definition.
What we do not claim: that each stop indicates a hazard, or that every protective stop was published.

Protective-stop concentration by location

What we observe: published protective-stop events and robot positions at those times.
What we derive: counts and rates by location, highlighting where recorded stops cluster.
What we do not claim: why stops cluster there, or that the location itself is problematic.

Findings card for protective stops in the month, by place, with the robots that stopped most — Product preview · synthetic test data

Protective stops by place with the rate per 100 hours, the robots with most, and the July count — Product preview · synthetic test data

5 · Change over time

What changed after a software update?

Before / after software updates

What we observe: published software versions or declared update times, plus recorded fleet behavior.
What we derive: changes in measurements across stated before/after windows, using the same rules and exposure basis.
What we do not claim: that the software update explains the observed difference.

Behavior drift over time

What we observe: the same published streams across successive periods, with configuration and coverage noted.
What we derive: trends in separation, speed, stops, and motion differences under consistent measurement definitions.
What we do not claim: that a trend will continue, or that it reflects a change in safety.

Findings card comparing the six updated robots with the other 24 after the Aug 12 software update, rate per 100 hours before and after under the same rules — Product preview · synthetic test data

Every change in the period with the finding rate before and after it, including a vendor software update on six robots — Product preview · synthetic test data

6 · Record quality

Findings card for record quality: the recorded share of the expected stream, gaps, undeclared robots, and clock offset — Product preview · synthetic test data

Completeness ratio

What we observe: received messages and available sequence or timing information within the declared capture scope.
What we derive: the recorded share of the expected stream, where an expectation can be established; the denominator is stated.
What we do not claim: a completeness percentage for streams whose expected volume cannot be established.

Gaps and missing streams

What we observe: breaks in sequence or timing, and expected streams absent from the captured record.
What we derive: gap windows, affected streams, and measurements limited by missing data.
What we do not claim: what happened during a gap, or whether missing data reflects the robot, network, or capture process.

Work with us

See what your fleet already publishes.

Start with the streams your fleet exposes. Together, we can establish which measurements the record supports.

What we measure

How close do robots get to people?

Closest approaches

Speed while people are present

Findings / Closest approaches

Fleet Safety Report / Closest approaches by place

Findings / Too fast near people

Fleet Safety Report / Too fast by place and shift