A discrepancy above threshold is noticed

What it is

Bend too hard and the floor feels wrung like a turntable; people stop or turn back to match a door. The detection threshold is the gain ceiling at which redirection still counts as “I am just walking.” Above it, the discrepancy is no longer absorbed by vision and is reported as a world that turns, a path that bends, or a shove from the system.

Below threshold, range expands. Above it, the event becomes a named disturbance. How range is expanded is a different leaf. This one only asks whether people noticed.

Why it happens

Spatial updating takes vision as the main channel, but proprioception and the vestibular system still vote. As gain rises, the arc under the feet, a centrifugal lean of the trunk, and a canal signal that fails to match virtual angular velocity during a real turn all leave the range that can be reinterpreted. Curvature gain often shows as an odd rotation of ground texture and as “walking straight while listing.” Rotation gain shows as the world overshooting or undershooting after a body turn, so a door no longer lines up.

The threshold is not one magic number. It moves with scene texture, walking speed, whether the path is being attended, and whether people were told “sometimes it bends.” Thresholds from two-interval forced choice are usually tighter than the “this feels wrong” threshold in free walking — forced choice is searching for a discrepancy; natural walking is trying to get there.

Studying it

The standard paradigm is a Steinicke-style two-alternative forced choice: walk a segment, ask whether the virtual path bent left or right of the physical one, and estimate a detection threshold with constant stimuli or a staircase. Curvature, rotation, and translation gains have to be measured separately; they do not collapse into one “redirection threshold.”

An ecological companion is also worth taking: no question, only spontaneous stops, spontaneous re-alignment turns, and spoken “it twisted.” Forced-choice thresholds bound the algorithm; spontaneous notice rates bound whether the experience has already broken. The roughly twenty-plus-metre curvature-radius figures common in labs are estimates at particular scenes and speeds, not device-independent constants.

Where it stops holding

Telling people “please find the bend” pushes the threshold down; what is measured is sensitivity under scrutiny, not under travel. In dark, low-texture scenes, or while a hand task is occupying attention, the threshold widens; an algorithm that pushes the wide value will show the moment people look up at the path. Vestibular-sensitive users have a narrower rotation-gain threshold. A reset is above threshold by construction; noticing during resets cannot be read back as the cruise threshold. Children, and different heights and gaits, change curvature geometry; adult lab thresholds do not transfer as-is.

Applying it

• Cap gain inside a detection threshold measured for the target users and speeds. Do not treat one paper’s median radius as a global constant.
• Keep cruise below threshold. When a reset is required, design it as an overt interruption, not as unnoticed redirection that failed.
• Tighten gain another notch in high-texture floors and long straight corridors, where the bend is easiest to see.
• How to check: without previewing “it will bend,” ask after the walk whether the path felt as if it turned or the world was wrung. Stable yeses, or turns back to align with a door, mean gain is supra-threshold — pull it down, do not make the arc prettier.

Related

• Same group: N4.05.1 Tiny rotational discrepancies expand the walkable range · N4.05.3 It only holds when the physical space is large enough
• Nearby: N4.04 Room Scale and Safety Boundaries · N4.09 Wayfinding and Landmarks in Space
• Search terms: redirected walking detection threshold · curvature gain JND · rotation gain