Robot safety: standards, sensing and validation

A complete robot-safety argument starts with the intended use and foreseeable misuse, identifies hazards throughout the lifecycle, estimates risk, applies risk-reduction measures, and verifies that the resulting system behaves as required. Mechanical limits, motion planning, control architecture, protective equipment, sensing, human procedures, and maintenance can all contribute.

Terms such as collaborative, compliant, soft, force-limited, autonomous, or covered in robot skin are not safety certifications. They describe possible design properties whose effect must be evaluated in the final application and environment.

The public ISO record describes ISO 10218-1:2025 as safety requirements for industrial robots and ISO 10218-2:2025 as the companion standard for industrial robot applications and robot cells. The ISO robotics overview also lists ISO/TS 15066:2016 for collaborative industrial robot applications.

Scope matters. The public ISO 10218-1 page lists exclusions, including service and medical applications. A humanoid used in a factory cell may intersect industrial-robot requirements differently from a service humanoid in a home, hospital, or public space. Determine the applicable edition, jurisdiction, product category, integrator responsibilities, and conformity route with qualified safety and legal professionals.

This page is a technical research map, not legal advice, a risk assessment, or a certification decision. The full paid standards and applicable local rules—not this summary—control formal compliance work.

Industrial robot safety often assumes a defined application, cell, tooling, workpiece, integration, and operating mode. Collaborative applications require evaluation of how people and the robot share a workspace. A mobile or humanoid platform can add locomotion, balance, batteries, whole-body contact, changing environments, general-purpose tools, learned behavior, and human proximity.

Those added capabilities broaden the hazard and validation surface. A humanoid benchmark can help make performance measurable, but performance benchmarking is not the same as demonstrating safety compliance.

External perception can monitor people and obstacles before contact. Joint torque, motor current, force-torque sensors, proximity sensors, bumpers, and surface tactile arrays can expose different parts of a collision or contact event. Their coverage and failure modes differ, so they may complement one another.

Robot skin is particularly relevant to distributed surface contact that a wrist sensor may not localize. But a research tactile array should not be described as a safety-rated protective device unless its claimed function, architecture, diagnostics, fault response, and validation satisfy the applicable requirements.

NIST’s Humanoid Robot Baseline Performance Benchmark project proposes low-footprint locomotion and manipulation tasks with quantifiable performance metrics and common comparison. It can improve measurement discipline and adoption readiness, but NIST does not present that project page as a safety certification scheme.

Research papers can establish sensor behavior or robot performance under named conditions. Manufacturer material can document intended features. A conformity assessment, application risk assessment, and validated safety function answer different questions. RoboSkin.ai labels these evidence levels instead of combining them into one safety claim.

Validation should trace each safety-related requirement to a method, acceptance criterion, result, and retained record. Software, model, payload, tooling, speed, environment, sensor, network, and operating-mode changes may invalidate earlier assumptions and need controlled review.