When we started designing Closer for mixed-traffic aisles, one of the first questions we had was: what does published research actually say about how workers behave around mobile robots in shared spaces? Not what designers assume, not what vendors claim in marketing materials, but what has been observed and measured in settings that resemble real warehouse and retail environments.
This post summarizes what we found and how it shaped decisions in the Closer motion design. We are drawing on published ergonomics, occupational safety, and human factors literature, not our own proprietary studies. We will note where our pilot observations aligned with or diverged from the published findings.
The primary behavioral response: avoidance and anticipatory adjustment
Multiple studies on human-robot coexistence in industrial settings consistently find that workers adapt their movement to anticipate robot positions, even when no instruction to do so has been given. A 2019 study from researchers at Chalmers University of Technology observed warehouse workers in a facility with SLAM-navigating AMRs and found that workers developed path-planning behaviors that accounted for predicted robot positions within the first two to three weeks of operation. Workers who had been in the facility for six months showed substantially more efficient path adjustments than workers who had been present for less than a month.
This finding has a practical implication: the first weeks of a deployment are the highest-friction period not because the robot is behaving differently, but because workers have not yet built a behavioral model of it. Safety briefings and supervisor observation during the adaptation period are more important than any change to robot configuration. The robot's behavior should be consistent during this period so that workers can build an accurate model.
Proxemics and minimum comfortable separation
Proxemics research (dating from Hall's 1966 work on interpersonal spacing norms) established that people maintain different separation distances from others depending on context. In industrial work settings, the concept of "personal space invasion" with robots has been studied more recently. Research from Waseda University's Human-Robot Interaction lab found that Japanese workers in logistics settings preferred a minimum robot-to-worker separation distance of approximately 0.6m to 0.8m in passing scenarios, measured at the point of closest approach. Below that distance, workers reported discomfort and showed avoidance behaviors including stopping and stepping back.
This finding directly informed our protection zone defaults. Our nominal zone boundary is set at 0.6m from the chassis edge in the direction of travel, which triggers deceleration rather than full stop. A comfortable passing clearance of 0.6m plus the half-chassis width of 0.35m gives a minimum comfortable lateral clearance of approximately 0.95m between robot path center and the worker's path center. For a 2.4m wide aisle, this leaves about 0.5m of aisle width beyond the required clearance, which is adequate but tight in high-traffic periods.
We are not saying 0.6m is the universal correct separation standard. Proxemics preferences vary across individual workers and cultural contexts. The published research skews toward European and East Asian industrial settings. We are saying 0.6m represents a reasonable design target for the facilities we operate in, and operators planning for higher-traffic aisles or workers with known sensitivity to proximity should consider wider aisle allocation for robot corridors.
The trust development arc and what disrupts it
Research on trust in human-robot collaboration consistently identifies predictability as the primary driver of trust formation. Studies from the MIT AgeLab and from Keio University's Robotics Research Group both found that workers increase their comfort level with AMRs over time not primarily because they come to believe the robot is "safer," but because they come to understand what the robot will do in a given situation. Predictability enables confident coordination.
The disruption factor is inconsistent behavior. Any change in the robot's behavior, even a change that is objectively an improvement, resets part of the trust development arc because it violates the behavioral model the worker has built. In one Keio study, a robot speed increase of 15% applied mid-trial caused workers who had previously shown fully adapted behavior to return to avoidance and monitoring behaviors for approximately one to two weeks before readapting to the new speed.
This has a direct implication for firmware updates and configuration changes in deployed units. If you are updating speed profiles, zone configurations, or deceleration tuning on a deployed robot, plan for a short re-adaptation period for workers and consider a brief briefing that explains what changed. What feels like a small software change to the engineering team may be a significant behavioral shift to someone whose work pattern has adapted around the previous behavior.
Worker stress and cognitive load in shared-aisle settings
Several occupational health studies have examined whether working in proximity to autonomous mobile robots increases worker stress levels. Findings are mixed and context-dependent. A 2021 study published in the journal Ergonomics examined pick workers at a mid-scale e-commerce fulfillment facility over a six-month period after introduction of AMRs, measuring cortisol levels and self-reported stress. Workers in aisles with AMRs showed elevated cortisol and higher self-reported stress in the first month, returning to baseline by month three.
However, a separate finding in the same study complicates the picture. Cognitive load measures, assessed through secondary task response time, showed that workers in AMR aisles had slightly lower cognitive load than workers in non-AMR aisles of the same facility. The researchers hypothesized that the AMR's presence as a moving reference point in the environment actually reduced the cognitive effort of spatial navigation for the workers, because the robot's predictable movement gave an additional spatial landmark in the aisle.
This is a counterintuitive result that we found genuinely useful. The implication is that robot motion should be predictable enough to function as a navigational reference rather than an unpredictable obstacle. A robot that moves in erratic paths, changes speed without apparent reason, or stops unexpectedly increases cognitive load. A robot with consistent, legible behavior patterns may, over time, reduce it.
Height, speed, and perceived threat
Human perception research on robot acceptance has repeatedly found that size and speed are the dominant factors in perceived threat from a mobile robot, with size having a somewhat larger effect. A tall robot moving at 0.5 m/s is perceived as more threatening than a low-profile robot moving at 1.0 m/s in controlled perception studies. This affects design choices that go beyond safety compliance.
For Closer's physical design, the chassis height profile was a deliberate decision: keeping the robot's maximum height at 1.2m rather than 1.6m (which would have been more structurally efficient for the payload bay) was partly an ergonomics choice. At 1.2m, the robot's eye line is below most workers' waists. It does not intrude into the face-and-upper-body space that proxemics research identifies as the highest-sensitivity personal space zone. We do not have data confirming that this choice improved worker acceptance in our specific pilots, but the published research provides a reasonable basis for the decision.
What the research does not resolve
Most published studies in this area are conducted in controlled conditions with small numbers of participants and short time horizons. Studies longer than six months are rare. Studies in actual production facilities with high throughput pressure and real job-performance consequences are rarer still. The ergonomics research base on human-AMR coexistence is growing but remains thin relative to the pace of AMR deployment.
Specific topics where we found very limited published evidence: the effect of robot density (multiple AMRs in the same aisle) on worker stress and adaptation time; worker adaptation differences by age and physical profile; the interaction between robot behavior and facility noise levels, which affects workers' ability to use sound as a proximity cue. These are gaps in the research base that affect real deployment decisions today.
Our approach is to treat the published research as informative design guidance rather than definitive prescription. The consistent findings, especially around predictability, adaptation time, and separation distance preferences, have directly shaped how we think about motion design and deployment support. The gaps remind us that there is no substitute for careful observation in each specific facility context, and that field observations in pilots will continue to be as important as laboratory research for informing how these systems should work in practice.
