Skip to content

Latest commit

 

History

2 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 

Repository files navigation

Robot Safety Standards — ISO 10218, ISO/TS 15066 & Global Compliance Guide

A comprehensive reference for robot safety standards worldwide, led by ISO 10218 (the binding international standard) and complemented by ISO/TS 15066, ANSI/RIA R15.06, and regional regulations. Includes practical guidance on Power and Force Limiting (PFL) verification and deterministic safety implementation.


Core Standards Overview

ISO 10218 — The Primary Standard

ISO 10218 is the international standard for industrial robot safety, published by ISO/TC 299 Robotics. It is a binding standard (not a technical specification) and all industrial robot installations must comply with ISO 10218 before being put into service.

  • ISO 10218-1:2011+A1:2020 — Safety requirements for the robot itself
  • ISO 10218-2:2011+A1:2020 — Safety requirements for robot system integration

Key principles of ISO 10218:

  • Safety functions must be deterministic and verifiable
  • Risk assessment must be performed for each application
  • Layered risk reduction (design → safeguarding → administrative measures)
  • Safety-rated software has specific development and validation requirements

ISO/TS 15066 — Collaborative Robot Add-On

ISO/TS 15066:2016 is a technical specification (guidance, not a binding standard) that adds collaborative robot safety requirements. It builds on top of ISO 10218 — you cannot comply with ISO/TS 15066 without first complying with ISO 10218.

Four collaborative operation modes:

  1. SMS — Safety-Rated Monitored Stop
  2. Hand Guiding — Human-operated control device
  3. SSM — Speed and Separation Monitoring
  4. PFL — Power and Force Limiting (most technically demanding)

Regional Standards

Region Standard Notes
International ISO 10218 Primary standard, globally recognized
USA / North America ANSI/RIA R15.06 Adopts ISO 10218 with national deviations
EU EN ISO 10218 Harmonized standard, CE marking relevance
Japan JIS B 8433 Japanese adoption of ISO 10218
China GB 11291.x Chinese national standard based on ISO 10218
Germany ISO 10218 + DGUV recommendations Additional BG/DGUV guidance

PFL Verification — The Hard Problem

Power and Force Limiting (PFL) is the most technically demanding collaborative mode. It requires demonstrating that contact forces and pressures stay within the limits defined in ISO/TS 15066 Annex A for different body regions.

Why PFL is Hard

  • Pressure depends on contact area — and contact area isn't a fixed number; it changes dynamically with force, material stiffness, and geometry
  • Body regions matter — the face has much lower limits than the leg; you need to know where contact happens
  • Static vs. transient — different limits for clamping (quasi-static) vs. impact (transient)
  • Dynamic contact area changes everything — most systems use fixed worst-case area, which is overly conservative but may still miss edge cases

Data: PFL Compliance Across 500 Robot Actions

Analyzing 500 robot action samples across 6 robot types and 3 scenarios reveals how often PFL limits are approached or exceeded:

Category Count Percentage Implication
PFL compliant (pressure < 150 kPa) 372 74.4% Safe for human contact per typical limits
Borderline / scenario-dependent 90 18.0% May be safe for some body regions, not others
Exceeds PFL limits (pressure > 300 kPa) 38 7.6% Unsafe for human contact — requires mitigation

Takeaway: A significant fraction of robot actions exceed typical PFL limits. Safety validation is not optional — it's a core engineering requirement for any collaborative or humanoid robot system.


Deterministic Safety — ISO 10218's Hidden Requirement

ISO 10218 doesn't explicitly say "deterministic," but the principle is woven throughout: safety functions must be verifiable, repeatable, and predictable. This is why traditional industrial safety uses hard-wired relays and PLCs, not software — let alone machine learning.

For modern software-based safety systems, this means:

  • No probability in the safety path — the safety decision must follow deterministic rules
  • Same input → same output — always, with no statistical variance
  • Traceable decisions — every safety decision can be audited and explained
  • No training data dependency — the safety system shouldn't depend on datasets

Why ML-Based Safety Doesn't Meet ISO 10218

Machine learning and VLA models are probabilistic by nature. They can be very accurate — 95%, 99%, even 99.9% — but:

  • 99.9% accuracy means 1 failure per 1000 checks
  • At 100Hz control rate, that's one unsafe decision every 10 seconds
  • Distribution shift can cause sudden drops in accuracy with no warning
  • You can't prove correctness — only measure performance on test data

This is why the correct architecture is hybrid: AI models handle perception and planning, while a deterministic physics safety layer provides the final ISO 10218-aligned safety gate.


Standards Compliance Checklist

ISO 10218-1 (Robot Manufacturer)

  • Risk assessment per ISO 10218-1 methodology
  • Protective stop functions (Category 0 and Category 1)
  • Speed and force limiting capabilities
  • Safety-rated software validated per IEC 61508 / ISO 13849
  • Deterministic behavior verified
  • Documentation, markings, and operating instructions

ISO 10218-2 (System Integrator)

  • Application-specific risk assessment
  • Appropriate safeguarding selected and installed
  • Workspace boundaries and access points defined
  • Emergency stop devices accessible from all positions
  • Operating instructions and user training provided
  • Initial safety validation and periodic re-validation

ISO/TS 15066 (Collaborative Operation)

  • Collaborative mode selected based on risk assessment
  • PFL: contact forces/pressures within Annex A limits for all body regions
  • PFL: contact area model validated (not fixed-area assumption)
  • SSM: separation distance formula correctly applied
  • Hand guiding: control device is safety-rated
  • SMS: stopping performance verified at all speeds
  • Collaborative workspace verification completed

Rotor Safety Engine — Standards-Aligned Implementation

Rotor Safety Engine

Rotor is designed with ISO 10218 principles from the ground up:

  • 100% deterministic — pure Newtonian mechanics, no probability, no ML
  • PFL verification — dynamic contact area + body-region-aware pressure calculation
  • 7-level risk granularity — progressive risk assessment, not just binary pass/fail
  • Semantic plausibility layer — first-line filter for impossible actions
  • Single file · zero dependencies — easy to validate and audit
  • ~17μs per check — fits in any real-time control loop
  • 349 test cases — comprehensive test coverage for validation

More Resources


⚠️ Disclaimer

This project is provided for research and educational purposes only. It is NOT a certified safety device, NOT a substitute for professional risk assessment, and NOT intended for use in safety-critical applications without independent validation.

The authors make NO representations or warranties of any kind, express or implied, about the completeness, accuracy, reliability, or suitability of the software. Any use is at your own risk.

Always comply with local safety regulations (ISO 10218, ISO/TS 15066, and applicable national standards) and conduct thorough risk assessments before deploying any robotic system.


The Deterministic Safety Gap

All current safety standards share one implicit assumption: safety is probabilistic. Safety Integrity Levels (SIL), Performance Levels (PL), and risk graphs all quantify the likelihood of harm — not the certainty of safety.

Rotor Dynamics introduces a complementary approach: deterministic physics safety.

Why Deterministic Matters

  • Standards say: "Reduce risk to an acceptable level"
  • Physics says: "This specific action at this specific force, speed, and contact area produces this specific pressure — here's whether it's safe"

The three core physical variables determine everything:

  • Force magnitude → contact pressure and tissue stress
  • Relative speed → impulse transfer and injury severity
  • Contact amplitude → affected area and energy distribution

When you can compute these deterministically from first principles, you don't need to estimate risk — you can calculate safety. This doesn't replace standards compliance; it makes it verifiable at runtime.

Validation Dataset

We maintain a growing dataset of 500+ labeled scenarios across seven risk levels, validated against biomechanical injury thresholds. The deterministic physics model achieves perfect consistency with physical measurement — because Newtonian mechanics doesn't have confidence intervals.

License

MIT — educational and reference use. Always consult official standards for compliance.