Introduction
Humanoid robots can create mechanical, electrical, software and cybersecurity hazards. A fall can injure a nearby person, a gripper can pinch or crush, a carried object can be dropped, and a remote connection can become another path for unsafe commands. Safety has to be assessed for the robot, its task, the workspace and the people who can enter that workspace.
ISO 10218-1:2025 covers industrial robots as machines and ISO 10218-2:2025 covers industrial robot applications and cells. Both standards explicitly exclude service robots and consumer products where the public has access. A humanoid used in a factory may fall within industrial machinery requirements, while a home or public-facing humanoid needs a different safety case and may involve other standards such as ISO 13482. No single certification claim should be treated as proof that every use of a humanoid is safe.
Key findings
- ISO 10218-1:2025 and ISO 10218-2:2025 apply to industrial robots and industrial robot applications, while their scope excludes service robots and consumer products used by the public.
- Risk assessment must include the actual task, payload, speed, reachable space, contacts, falls, maintenance and foreseeable misuse.
- Learned policies should operate inside independent limits for motion, force, workspace and stopping rather than carrying the entire safety function themselves.
- Remote supervision adds network, authentication and command-integrity risks that need controls separate from motion planning.
- Public videos rarely disclose fault tests, stopping performance, intervention logs or the exact scope of any certification.
Humanoid robot safety evidence by operating context
The table records what each source establishes and keeps missing data visible.
| System or method | What the evidence establishes | Evidence class | Main unresolved point |
|---|---|---|---|
| Industrial humanoids | Need application-specific integration with industrial robot and machinery safety requirements. | System-level safety case | No single standard covers every humanoid application. |
| Home humanoids | Operate around children, pets, stairs, heat and privacy-sensitive spaces, often outside established industrial cells. | Emerging safety challenge | Certification claims must identify product, version, standard and scope. |
| Research humanoids | May be operated in controlled labs with trained staff and restricted workspaces. | Controlled research risk | Public demonstrations reveal little about fault testing. |
| AI policy safeguards | Can reduce unsafe action proposals but do not replace independent motion and electrical protection. | One safety layer | No single standard covers every humanoid application. |
Evidence classification
- Officially documented specifications, standards or project status stated by the responsible organization.
- Real-system evidence demonstrations or deployments performed on physical hardware under described conditions.
- Company claim a numerical or operational statement reported by the company and not independently audited.
- Simulation or research evidence useful for mechanisms, but not proof of field deployment.
- Insufficient public evidence control mode, trial count, version or operating conditions are missing.
Start with the application and reachable hazards
A safety review begins with the specific job. Record robot mass, payload, maximum speed, joint torque, reachable space, end-effector geometry, fall envelope and the people who can enter the area. The same humanoid can have a different risk profile when moving empty totes, using a tool or working near a child.
Foreseeable misuse also belongs in the assessment. Examples include a person entering during recovery, an operator carrying a load outside the intended range, a sensor becoming occluded or a remote user issuing a command after the scene has changed.
Industrial standards have defined scope
ISO 10218-1:2025 specifies safety requirements for industrial robots. ISO 10218-2:2025 covers integration, commissioning, operation, maintenance and decommissioning of industrial robot applications and cells. Both standards exclude service robots and consumer products where the public has access.
That scope distinction matters for humanoids because a factory machine and a home robot can share a similar body while operating under different conditions. A certification statement should name the product version, the standard, the assessed function and the operating context.
Independent limits should contain learned policies
A learned controller can produce a motion that is valid for the policy and still unsafe for the current contact, payload or workspace. Safety-related limits can cap speed, force, torque, joint travel and reachable zones independently of the high-level policy.
Emergency stopping, protective stopping and safe-state behavior also need defined paths and reset rules. For a biped, removing actuator power immediately can itself create a fall hazard, so the safe response depends on the machine and application.
Perception and balance failures need physical tests
Humanoids can miss transparent objects, occluded people, low obstacles or unexpected contact. A useful test plan changes lighting, object placement and approach direction, then records detection failures, contact events, falls and recovery behavior.
Testing should include sensor loss, stale observations, communication faults and changes introduced by software updates. A successful task video does not reveal these fault responses.
Remote supervision adds cybersecurity requirements
Teleoperation and remote recovery create a control path between a networked system and physical actuators. Authentication, access control, command integrity, logging and network segmentation therefore become part of the safety case.
NIST SP 800-82 Rev. 3 provides guidance for operational technology security, including risk management, architectures, threats and security controls. It is general OT guidance rather than a humanoid certification standard.
Evidence to request before a deployment
For a factory pilot, ask for the applicable standards, risk assessment, stop architecture, tested payload and speed ranges, fault-response tests, intervention logs and maintenance procedure. For a home or public robot, also ask how the system handles stairs, children, pets, privacy-sensitive cameras, remote access and loss of connectivity.
Public evidence is still incomplete for many humanoid products. When a manufacturer publishes a safety claim without the tested standard, product version or operating scope, the missing fields should stay undisclosed rather than being inferred from a demonstration.
Limitations and missing information
- No single standard covers every humanoid application.
- Certification claims must identify product, version, standard and scope.
- Public demonstrations reveal little about fault testing.
- Specifications, prices, repositories and deployment status can change after publication.
- Benchmarks from different robots or environments are not directly comparable.
Conclusion
Humanoid safety has to be demonstrated for a named robot, task and operating environment. Industrial deployments can draw on ISO 10218 requirements where the robot and application fall within scope. Home and public-facing systems require a safety case that addresses public access, falls, contact, remote assistance and privacy. Buyers and operators should ask for the exact standard, tested configuration, fault responses and intervention data behind each safety claim.
Frequently asked questions
What makes a humanoid robot safe?
Safety depends on the robot and the application. Relevant controls can include mechanical design limits, speed and force limits, collision detection, balance and fall handling, safety-related stopping, supervision, maintenance procedures and secure remote access.
How should humanoid robot safety be evaluated?
Document the task, payload, speed, reachable space, end effector, people at risk and foreseeable misuse. Then verify the safeguards, stop behavior, fault tests, intervention procedure and the exact scope of any claimed standard or certification.
Which standards apply to humanoid robots?
ISO 10218-1:2025 covers industrial robots and ISO 10218-2:2025 covers industrial robot applications and cells. Their published scope excludes service robots and consumer products where the public has access. Other applications can involve standards such as ISO 13482 and jurisdiction-specific machinery or product-safety requirements.
What information is often missing from public safety claims?
Public material often omits the assessed product version, standard scope, stopping performance, test payload, fall tests, fault injection, intervention counts and long-duration operating data. Those fields should stay undisclosed when the manufacturer has not published them.
Can a humanoid be safe while using a learned policy?
A learned policy can operate inside a safety architecture that imposes independent motion, force, workspace and stopping limits. The required design depends on the application, and the policy should not be assumed to provide every safety-related function.
Sources and methodology
Sources were rechecked on September 26, 2026. The review uses the published scopes of ISO 10218-1:2025 and ISO 10218-2:2025, ISO robotics standards information and NIST SP 800-82 Rev. 3 for operational-technology security. Product claims remain attributed to the manufacturer, and unpublished test results remain undisclosed.
Related TechniaHQRobot guides
Source review notes
Review date recorded September 26, 2026
Supported by the cited sources
- ISO 10218-1:2025 covers industrial robots and ISO 10218-2:2025 covers industrial robot applications and cells.
- The published scopes of ISO 10218-1:2025 and ISO 10218-2:2025 exclude service robots and consumer products where the public has access.
- NIST SP 800-82 Rev. 3 provides cybersecurity guidance for operational technology systems.
Not confirmed or incomplete
- Public product pages rarely publish enough fault-test data to compare humanoid safety across manufacturers.
- Certification claims need the assessed product version, standard and scope before they can be applied to a specific deployment.
Likely to change quickly
- Robot software versions, safety features, deployment conditions and certification status.
- Commercial availability and product configurations.
Share this article
Share the current TechniaHQRobot article page.
Continue reading
Open the latest robotics reporting, Physical AI analysis and hardware notes.