How safe are humanoid robots?
Safety is one of the key blockers to the mass adoption of humanoid robots—despite impressive progress from Tesla’s Optimus or Figure’s BMW partnership.
A major hurdle is the lack of established standards that regulatory bodies like OSHA or courtrooms can refer to in liability cases. We believe that pre-emptive investment in safety analysis and preparedness will allow humanoids to reach market readiness faster than their predecessors like AMRs or drones.
1. Understand Your Context: The State of Standards
The development of functional safety standards for humanoids is in flux. Recent efforts—like those from ASTM Subcommittee F45.06 on Legged Robot Systems—show promise.
Key developments include:
- WK86916: New test methods for disturbance rejection in legged robots
- Evaluation practices like terrain perturbation (e.g. rough surfaces, shifting platforms, tilted treadmills)
Agility Robotics' Digit and efforts from leaders like Aaron Prather (IEEE RAS) are shaping the field.
Takeaway: The most important action today is collecting robust reliability data for future standard alignment—well beyond traditional ISO 10218 or ISO 12100 metrics.
2. Functional Safety Parallels and Divergences from CoBots
Much of humanoid safety follows conventional approaches:
- ISO 13849: Machinery safety
- ANSI/RIA R15.06 and R15.08: Robot and AMR safety
- Risk frameworks covering pinch, crush, shear, entangle, and impact hazards
Where humanoids diverge is in legged locomotion and fall risk.
While AMRs use techniques like trajectory deviation shutdowns, legged robots require stability recovery mechanisms that these rules can’t directly apply to.
Designers must also implement redundant safety control systems that can override high-level autonomy—leveraging robust sensors like LiDAR/RADAR, not just internal state feedback.
Saphira is leading efforts to make functional safety engineering approachable for humanoid development.
3. Innovations in Safety Technology
Modern safety in robotics is advancing rapidly across several fronts:
Complex Sensor Redundancy
- Affordable safety-rated sensors: Velodyne, Luminar, SICK
- Systems like 3Laws integrate these into structured safety architectures
- Need for certified vision-based safety systems is high—Saphira is actively collaborating on this
High-Speed Safety Processing
- Platforms like Nvidia IGX Orin and SICK safeVisionary2 offer fast, PLC-compliant computation
- Enable real-time safety logic at the edge
Wireless Safety Communication
- FORT Robotics enables fast, secure wireless safety links
- Opportunity: Open-source safety communication protocol for broad adoption
Human Presence Detection
- Advanced sensing: ultrasonic, capacitive, radar-based (e.g. Inxpect)
- Governed under IEC 62998, with IEC 62998-3 starting to standardize AI-in-safety applications
Perception for Rule-Based Control
- Binary-safe outputs (e.g. “human detected?”) now feasible
- Combine modern perception stacks with compliance to ISO and IEC rules
4. Global Safety Standards Shaping Humanoids
Humanoid robots intersect with multiple standard domains:
- ISO 13482: Personal care robots
- ISO 10218: Industrial robotics
- ANSI/RIA R15.08: AMRs
- JIS B 8445 (Japan), CSA Z434 (Canada)
5. Related and Emerging Frameworks
UL 4600, ISO 26262, ISO 21448
These vehicle-focused standards offer:
- System-level hazard assessment
- Safety function independence
- Monitoring of joint torques, trajectories, and ground contact
- Edge-case handling, like slip recovery or surprise human interaction
IEC 61508
Applies to electrical/electronic systems, informing sensor and control reliability.
BSI 8611
Covers ethical robot deployment—critical for public-facing humanoids.
6. Academic, Policy, and Industry Contributions
Research
- IEEE RAS / ICRA / Humanoids Conference
- TU Delft and other EU labs
- NIST test methods for performance benchmarking
Policy
- WEF reports on AI & robot governance
- UNECE work on intelligent mobility standards
Industry
- Boston Dynamics, Agility Robotics: Pushing real-world implementations
- Tesla: Exploring humanoid robot regulation
- SICK, Velodyne, Luminar: Safety component innovation
7. Conclusion: Preparing for a Safe Future
As humanoid robots integrate into the workplace and public life, functional safety will determine their success. Whether through legged stabilization, redundant control, or modern perception, robots must meet increasingly rigorous standards.
At Saphira AI, we’re working with robotics companies to make these standards practical, accessible, and proactive.
👉 Contact us to discuss how we can help you prepare your systems for both current and future certification frameworks.
Special thanks to Zhaoyuan Gu (Georgia Tech) for contributions to this article.
