Nearfield starts from evidence people can inspect, not a marketing claim.
Sample report / NF-RR-001
A robot demo is not proof that it can work around people.
Nearfield turns a robot clip into a readable deployment decision: what happened, where humans felt uncertainty, what risk remains, and what the team must fix before a real pilot.
Plain-English resultIt works once. It is not trusted yet.
Evidence clipCorridor delivery / 00:19
00:07Human slows first
00:11Late pause
The report marks the exact moments where the robot becomes hard to read.
The output is a clear deployment decision, constraints, and fixes.
Indoor delivery platform in a public corridor.
Navigate through shared space and reach a handoff point.
Close enough for comfort and intent clarity to matter.
The task succeeds, but the human interaction needs repair.
What normal users should understand
A robot demo is not the same as deployment proof.
A robot can finish a task and still fail the human-space test. The important question is not only whether the robot reached the goal. It is whether nearby people understood it, felt comfortable around it, and knew what would happen next.
Evidence timeline
Where the demo stopped being deployment-grade.
00:03Watch
Robot enters shared pathThe robot starts moving into the same corridor lane a person is about to use.
00:07Risk
Human slows down firstThe person adjusts to the robot before the robot clearly yields.
00:11Risk
Robot pause arrives lateThe robot pauses, but only after the human has already changed pace.
00:14Fix
Task resumesThe robot continues, but there is no visible signal that the conflict is resolved.
00:19Done
Delivery completedThe task succeeds, but the interaction quality is not deployment-grade.
Readiness scores
High task success, weak human confidence.
Task completionA-
86The robot finished the delivery task.
Motion legibilityC
62People could not easily predict the next move.
Human comfortC-
58The robot came too close before yielding.
Timing controlC+
64The slowdown happened later than it should.
Fallback behaviorD
41The robot did not show a clear recovery state.
Scene fitB
73The behavior is acceptable only in a controlled route.
Failure modes
What went wrong?
Late deceleration near a human crossing path.
No visible intent cue before entering the shared lane.
The human adapted to the robot instead of the robot adapting to the human.
No clear stop-and-yield fallback when social uncertainty appeared.
Human interaction riskMedium-high
The person changes pace and body direction because the robot's next action is unclear. That means the human is carrying the safety burden, not the machine.
Deployment constraints
Where this robot may operate today.
Supervised demo
Controlled indoor route
Low-density traffic
Unsupervised public deployment
Crowded retail environments
Child-heavy or elderly-heavy spaces
Recommended fixes
What the robot team should change before the next run.
The robot should decelerate before entering a human path, not after the person reacts.
Add a visible cue before turning, crossing, yielding, or resuming movement.
When a person enters first, the robot should stop, hold, and make the next action obvious.
Run the same task with at least five human approach angles before upgrading the verdict.
Not ready for open human environments.
Next milestone: pass the shared-path interaction test with no human hesitation events and a visible yield state before route conflict.
Submit a robot demo