A robot should behave cautiously when a person is in a blind spot

What it is

Blind-spot-aware cautious behaviour slows, separates, changes viewpoint, confirms, or stops when the robot cannot establish that a region is clear. “No person detected” is distinct from evidence that nobody is present.

Why it happens

Treating unknown cells as free converts sensing gaps into traversable plans. Cautious planning assigns unknown occupancy a cost and seeks a new view; slowing creates detection and braking margin. Excess conservatism can immobilise the robot in structural blind spots, so safe information-gathering action is needed.

Studying it

Occlusion, target appearance, and policy can be varied using near misses, task time, information gain, unnecessary stops, and comprehension. Surrogates and simulation cover extremes before low-speed hardware. Incident rate alone can hide an unusable always-stopping policy.

Where it stops holding

Other verified isolation can establish that a static region is empty. Active sensing can itself intrude. Risk analysis defines the minimum; user confirmation does not replace independent protection, and communication loss keeps the conservative state.

Applying it

• Preserve free, occupied, and unknown as distinct map states that affect speed and route.
• Prefer sensing moves that do not enlarge swept space; otherwise stop outside and ask an authorised person.
• Test sudden appearance, low targets, and moving occlusion, combining near misses, stop count, recovery, and correct explanation.

Related

• Same group: X3.06.1 Robot sensing is not omnidirectional, so blind spots must be known · X3.06.3 Physical indication of sensor orientation supports on-site judgement better than documentation · X3.06.4 Sensor fusion can mask component failure without eliminating system blind spots
• Adjacent: X3.08 Communicating uncertainty · X5.05 Responding when a person enters the workspace
• Search terms: unknown space · cautious motion planning · active perception