Data-Driven Configuration of Protective Separation Zones in Collaborative Robot Cells from Operator Occupancy

Aulon Bajrami, Theo Jacobs, Werner KrausarXiv 2026 · arXiv

TL;DR

This paper introduces a data-driven method to optimize the safety zones of collaborative robot cells by analyzing actual operator movement patterns. By replacing "worst-case" safety assumptions with real-world occupancy heatmaps, the researchers achieved a 17% reduction in safety field size and a 9% boost in productivity without compromising safety standards.

What Problem Does This Paper Address?

In collaborative robotics, safety is governed by the "Speed and Separation Monitoring" (SSM) principle (ISO/TS 15066). Currently, engineers design protective separation zones based on a theoretical "worst-case" scenario—assuming an operator could be anywhere at any time.

This conservative approach leads to two main inefficiencies:

  1. Oversized Zones: Large safety fields cause robots to stop frequently due to "unrelated traffic" (people just walking past the cell).
  2. Reduced Speed: To keep zones small, robots are often forced to move at lower nominal speeds to ensure they can stop in time if someone enters the field.

Furthermore, traditional risk assessments are often subjective and time-consuming. There is a clear industry demand for objective, data-driven evidence to support safety configurations.

Methods

The authors developed a non-safety-rated "edge camera unit" that can be retrofitted to existing robot cells. The system follows a multi-stage pipeline:

  1. Perception: Using an RGB-D camera (Intel RealSense), the system employs Lightweight OpenPose to track human skeletons in 2D. These are then "lifted" into 3D using depth data and projected onto the floor plane.
  2. Activity Recognition: The system distinguishes between operators interacting with the cell and people simply passing through. This prevents "nuisance" data from skewing the safety model.
  3. Occupancy Heatmaps: Over several shifts, the system aggregates positions into a heatmap showing where operators actually spend their time.
  4. Optimization: The core of the paper is an advisory algorithm that evaluates different candidate safety zone radii ($R$). It solves a cost function that balances the "cost" of protective stops (caused by large fields) against the "cost" of reduced robot velocity (required by small fields).
Overview of the proposed data-driven approach

Crucially, the system is advisory. It does not control the robot directly; instead, it provides a proposal for a safety engineer to review and incorporate into a formal risk assessment.

Datasets

The method was evaluated in:

  • Laboratory Environment: Initial testing and validation.
  • Industrial Beta Tests: Two operational robot cells at different industrial partner sites.

Due to non-disclosure agreements, the raw trajectory data is not public, but the results are reported in an aggregated, anonymized format.

Key Insights and Ideas

  • Spatial Imbalance: The heatmaps revealed that operator presence is usually concentrated in very compact regions (e.g., in front of a fixture), while large portions of the "worst-case" safety zones are almost never entered.
  • The Field Size Paradox: A smaller safety field isn't always better. While it reduces accidental triggers from passers-by, it forces the robot to move slower (or brake harder) to satisfy the physics of stopping distances. The "sweet spot" depends entirely on the specific traffic patterns of that factory floor.
  • Separation of Concerns: By keeping the perception system "advisory" and outside the safety-rated signal path, the authors made the system much easier to deploy in real factories. It provides evidence for humans to make decisions rather than trying to replace certified safety hardware.
Occupancy heatmaps showing concentrated operator presence

Key Results

The beta tests yielded significant improvements:

  • Field Reduction: The protective safety distance was reduced by approximately 17% (e.g., from 1.744 m down to 1.435 m in one cell).
  • Productivity Gain: Average productivity increased by 9% across the instrumented cells.
  • Optimization Visualization: The system successfully mapped the trade-off between robot velocity and safety distance, allowing engineers to see exactly where the current configuration was leaving performance on the table.
Optimization of robot velocity vs. safety distance

Limitations and Future Work

Despite the technical success, the authors noted significant human and organizational hurdles:

  • Risk Assessment Inertia: Companies are often reluctant to modify a safety concept that has already passed inspection, as any change requires a formal (and often expensive) re-assessment.
  • Privacy Concerns: Even with anonymized skeleton tracking, industrial partners expressed reservations about camera-based monitoring on the shop floor.
  • Deployment Timing: The authors suggest this method is most effective during the planning and commissioning phase rather than as a retrofit, to avoid the "re-certification" headache.

Future work aims to include more perception modalities, such as gaze tracking (to see if the operator is paying attention) and ergonomic monitoring (using RULA scores), to provide an even richer data set for risk assessment.

Wider concept for data-supported observation

Summary updated

Abstract

The protective separation zone of a collaborative robot cell is commonly dimensioned for a worst case that disregards where the operator is actually present during the task. This paper presents a camera-based, data-driven method that configures this zone from observed operator occupancy. Operator positions accumulated over several production shifts are aggregated into an occupancy heatmap, from which a protective field is derived that trades off the achievable robot velocity against the required safety distance, following the speed and separation monitoring principle of ISO/TS 15066. The resulting configuration is advisory, that is, reviewed within the risk assessment rather than applied autonomously to the robot controller. The method, which builds on a pose-based perception pipeline with activity recognition, was evaluated in two operational cells at two industrial partners, where the protective field could be reduced by approximately 17% and a productivity increase of 9% was measured. However, adoption proved to depend on more than the demonstrated gain alone.
BibTeX
@misc{bajrami2026data,
  author = {Aulon Bajrami and Theo Jacobs and Werner Kraus},
  title = {{Data-Driven Configuration of Protective Separation Zones in Collaborative Robot Cells from Operator Occupancy}},
  year = {2026},
  eprint = {2611.07621},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.07621},
}

Loading related content...

Figure