From Machinery Safety to Industrial Humanoids: Why the Next Era Requires More Than Traditional Standards
This article traces the evolution of machinery and industrial robot safety, from mechanical safeguarding to functional safety, autonomy, and AI-driven behavior. It examines why traditional safety standards remain essential but are no longer sufficient on their own, how emerging standards address new challenges, and why humanoids and other Physical AI systems require a layered approach to safety.
It was written by Jody Nelson, SRES Managing Partner and Co-Founder, a veteran safety leader with more than 25 years of experience advancing safety-critical automotive and autonomous systems for OEMs, Tier 1/2 suppliers, semiconductor companies, and organizations developing autonomous vehicles, robotics, and other Physical AI systems.
Looking to go deeper? Explore SRES’ Physical AI Training, including ISO/IEC TS 22440 Functional Safety and AI Systems Training. Need support beyond training? Explore our Physical AI consulting services, including robotics safety.
Introduction
When I was in high school, I hated history. I never felt connected to it, and it always seemed like a collection of events that belonged to someone else. As I have grown older, my perspective has changed. After working in safety for twenty-five years, I now see my own experiences reflected in the evolution of the field. The standards I have worked with, the technologies I have watched emerge, and the failures and successes I have lived through are now part of the history that is shaping the next generation of robots and humanoids.
For nearly fifty years, industrial safety has evolved through a steady and predictable progression. The field began with mechanical safeguarding. It moved into functional safety. It expanded into robotics. Today we are entering a phase that the traditional standards were never designed to handle. Robots and humanoids now rely on autonomy, perception, and machine learning.
A clear inflection point appears in the 2005 to 2006 timeframe. This was the moment when the industry shifted from purely mechanical protections and hardwired circuits to safety rated electronics, software, and probabilistic risk modeling. Safety engineering became a discipline of systems rather than a discipline of machines.
The Foundations: Machinery Safety Built the Discipline
I could start this history much earlier. Power presses, stamping machines, and early factory legislation all played important roles in shaping machinery safety. However, I am choosing to begin in the mid-1970s, which conveniently aligns with my own entrance to earth. This is also the period when the United Kingdom introduced BS 5304, a document that effectively codified the beginning of what we consider modern machinery safety. It established structured safeguarding principles, consistent terminology, and a repeatable way to evaluate hazards. In many ways, BS 5304 is the starting point for the discipline we practice today.
The early standards such as BS 5304, EN 954-1, and the first editions of ISO 13849 were built for a world of physical guards, interlocks, and relay logic. EN 954-1 evolved directly from BS 5304 and carried its safeguarding concepts into the European Machinery Directive. When ISO sought to create a global machinery safety standard, EN 954-1 became the foundation for the first edition of ISO 13849-1. This progression established the architecture categories and qualitative design principles that shaped machinery safety for the next decade.
By the late 1990s and early 2000s, IEC 61508 and ISO 13849-1 (2006) pushed safety into the electronic and software domain. Performance Levels, SILs, diagnostic coverage, and fault tolerance became part of everyday engineering vocabulary. These standards did more than reduce risk. They professionalized the discipline.
This foundation still matters. Without it, the robotics and autonomy boom of the 2010s and 2020s would have been chaotic.
Robotics Safety: From Fences to Functional Safety to Autonomy
Industrial robot safety followed a similar path. The first robot standards such as ANSI/RIA R15.06 (1986) and ISO 10218 (1992) focused almost entirely on keeping people out of the workspace. Fences, interlocks, and lockout procedures were the entire safety strategy.
By the early to mid-2000s, robots were no longer standalone arms. They were integrated into highly complex, multi-machine robotic systems that required coordinated controls, shared safeguarding zones, and system-level hazard analysis. This shift created new challenges that the earlier standards were never designed to address.
By 2006, robotics entered the functional safety era. ISO 10218 finally recognized that safety certified software and electronic controls could be trusted to monitor speed, position, and separation zones. Collaborative robots arrived soon after. Mobile robots followed with ISO 3691-4 and ANSI R15.08 in 2020.
However, these standards still assumed something important. They assumed that the robot’s behavior was deterministic.
If a sensor failed, the failure modes could be enumerated. If a controller faulted, the safe state could be predicted. If the robot stopped, it stayed stopped.
Humanoids and modern autonomous robots break these assumptions.
The New Reality: Nondeterministic Systems and AI-Driven Behavior
Traditional functional safety handles deterministic failures. Autonomy introduces nondeterministic failures. These failures include misclassification, Simultaneous Localization and Mapping (SLAM) corruption, planning instability, blind spots, and behavior that varies with lighting, clutter, reflectivity, or edge case environments.
A modern robot is a complex system. It combines perception, planning, mobility, manipulation, and software-driven decision making. Standalone machinery safety standards such as ISO 13849 or IEC 62061 are not sufficient to evaluate or mitigate the risks of the complete robot. They are designed to assess individual safety functions, not the behavior of an integrated robotic system operating in a dynamic environment. This is why system level functional safety is required, and why IEC 61508 becomes essential for robots that rely on autonomy and AI.
This shift is also why the standards landscape is changing.
Examples include the following, which at the time of this publication have not yet been released:
- ISO 26058 for statically stable robots
- ISO 25785 for dynamically stable robots such as humanoids
- ISO/IEC TS 22440 for AI-enabled perception and decision making
- The EU Machinery Regulation which requires Notified Body review for ML-based safety components, legally mandated in Europe
- ISO 12100 (2nd Edition) which will explicitly address unintended self-developing behavior
- IEC 61508 (3rd Edition) which will acknowledge nondeterministic algorithms for the first time
These changes are structural. They are not incremental updates.
Why Traditional Standards Still Matter, but Are Not Enough
The machinery and robot safety standards give us the backbone. They provide hazard identification, risk reduction, safe states, redundancy, diagnostics, traceability, and verification. These principles remain essential.
However, autonomy and AI introduce failure modes that cannot be fully enumerated and do not behave consistently. You cannot solve misclassification with a redundant sensor. You cannot solve SLAM corruption with a safety relay. You cannot solve planning instability with a PL-rated controller.
When discussing these limitations, many safety engineers point to supplementary industrial standards like IEC TS 62998, which was introduced to address sensor performance limits and environmental triggering conditions. IEC 62998 provides a structured way to ensure a system can reliably perceive its environment, covering SOTIF‑like aspects for the sensing domain, but it stops at perception. It ensures the robot can accurately detect a hazard, but it does not evaluate how nondeterministic behavior from AI decides to react to that hazard.
In addition to nondeterministic failures, autonomy also creates hazards in situations where the system behaves exactly as designed without any component faults. This is where ISO 21448, known as SOTIF, becomes essential. Originating in the autonomous vehicle space, SOTIF focuses on scenarios where the robot is “safe but confused” by an edge-case environment. A sensor might be temporarily blinded by a reflection, or a perception algorithm might misinterpret an unusual shadow. The planner then generates a mathematically valid trajectory based on this flawed understanding, and the robot executes the motion precisely as intended, resulting in a hazardous situation. For robots and humanoids operating in unstructured or dynamic spaces, these SOTIF-type hazards are unavoidable and must be addressed explicitly.
This is why the future of industrial robot safety, especially for humanoids, requires a layered approach.
- Quality and safety management systems (ISO/IEC 42001 and IEC 61508)
- AI and safety risk management (ISO 12100 and ISO/IEC TS 22440)
- Functional safety (IEC 61508 and ISO 13849)
- Physical AI and autonomy safety (ISO/IEC TS 22440 and ISO 21448)
- Product specific safety (ISO 10218, ISO 3691-4, ANSI R15.08, ISO 26058, ISO 25785)
Traditional machinery safety is the foundation. It is no longer the ceiling.
The Takeaway: Robots Are Becoming Physical AI Systems
Industrial robots used to be deterministic machines. Humanoids and advanced mobile manipulators are becoming physical AI systems.
The safety discipline must evolve accordingly.
We are entering an era where functional safety is necessary but insufficient. AI safety and autonomy safety must be integrated. Standards must address nondeterministic behavior. Notified Bodies will evaluate ML-based safety components. The boundary between robot safety and AI safety is disappearing.
The companies that succeed in this transition will be the ones that treat autonomy and AI safety as first-class engineering disciplines rather than add-ons to traditional machinery safety.
Have insights or questions? Send us an email at info@sres.ai or leave a comment below—we welcome thoughtful discussion from our technical community.
Interested in learning more about our approach? Explore why teams choose SRES training and how we support organizations with consulting for Automotive and Physical AI applications.
