Functional Safety for Humanoid Robots

How safe are humanoid robots?

Safety is one of the most important blockers to mass adoption of humanoid robots, which are starting to achieve sufficient functionality to be useful in various environments, like Tesla’s deployment of Optimus in its factories, and Figure’s partnership with BMW!  

Part of the story here is that there aren’t baseline or composite standards references that OSHA (The US Occupational Safety and Health Administration, which sets regulations for workplace safety) and its peers, as well as courtrooms, can use to gauge liability in failure cases.  

We believe that pre-emption of preparedness for this sort of analysis and robustness of safety standards can ensure that humanoids reach the market faster than their predecessors, AMRs and commercial drones.

The development of functional safety standards for humanoid robots is undergoing a transformative phase, driven by advancements in robotics technology and a growing emphasis on human-robot collaboration. 

Recent discussions within ASTM Subcommittee F45.06 on Legged Robot Systems have yielded promising progress, particularly in defining disturbance rejection testing methods for legged robots. 

One notable outcome is WK86916: New Test Methods for Disturbance Rejection Testing of Legged Robots, which establishes a standardized approach to evaluating a legged robot's ability to recover from external forces.  The other common approach for evaluation, which hasn’t yet been standardized, is terrain perturbation, such as making the robot walk on rough terrain, on a tilted treadmill, or even on shifting platforms, as described in the paper “From the State of the Art of Assessment Metrics Toward Novel Concepts for Humanoid Robot Locomotion Benchmarking”. 

Agility Robotics’ Digit serves as a compelling example of these capabilities, and leadership from industry experts such as Aaron Prather continues to push the boundaries, including through efforts within the IEEE RAS Humanoid Robotics Study Group. 

Inevitably, kick test-proof stabilization is the key function that most legged robot producers are focused on innovating on, and many vendors like MAB Robotics offer IMUs and actuators that make it easy to implement these functions. 

To prepare for certification in the future, the most important action to be taken right now is effective reliability data collection so that performance metrics are easily accessible when standards are published and common reports are expected by customers, extended beyond the traditional metrics dictated by ISO 10218-1 and 12100.  

Metrics monitoring and risk assessment are key capabilities that Saphira AI can help you with!  Contact us for more information!

Functional Safety in Humanoid Robots: Parallels and Divergences from CoBots

For the most part, functional safety assessments for humanoid robots follow a well-trodden path shared with other human-cooperative robots. 

Standards such as ISO 13849 (Safety of Machinery) provide a framework for evaluating hazards like pinching, crushing, shearing, cutting, entangling, and impacts. 

The familiar methodologies align well with the risk assessments employed in industrial and autonomous mobile robots (AMRs), such as those outlined in ANSI/RIA R15.06 and R15.08. 

However, humanoid robots diverge in a crucial area: legged behavior.

Unlike wheeled robots, legged robots must address the critical hazard of falls. 

Standards for AMRs, such as collision sensing and trajectory deviation rules (e.g., shutting off after a 15 cm deviation), provide inspiration but must be adapted to the unique dynamics of humanoids. 

The challenge lies in ensuring stability and recovery in environments where traditional rules for navigation and safety are insufficient.  

This gap is one of the most important reasons humanoids aren’t widely deployed in factories yet.

Despite this focus on legged behavior, the most important facet of safety to keep in mind is mitigation of the common hazards in the particular domain of tasks your robot is sold for: you, as the producer of a humanoid robot, must include a safety controller that handles inputs regarding possible stops, ideally from duplicated and more robust sensor inputs than standard processing, such as LiDAR or RADAR.  

We at Saphira think this is an enormous area of upcoming work, where we’re leading the pack on making the traditional facets of functional safety engineering straightforward for humanoid development!

Progress and Innovations in Safety Technologies

The landscape of functional safety for humanoid robots has significantly evolved thanks to innovations in sensing, processing, and communication technologies. Many of the challenges identified in early scoping papers, such as James H. Graham's 2000 IEEE paper presented at the MIT Humanoids Conference, are now being addressed:

1. Complex Sensor Redundancy
   Safety-critical redundancy in sensors like LiDAR, previously expensive and niche, is now more accessible with solutions from Velodyne, Luminar, and safety-rated manufacturers like SICK. Advanced systems like 3Laws have emerged to integrate these commodity sensors into reliable safety architectures. There is a huge need for certified vision-based safety systems that are a key focus in industry now, on which we are actively collaborating with various robotics companies!

2. High-Speed Processing for Safety
   The demand for fractional-second response times and ultra-low probabilities of Type 1 errors is met by modern hardware. High-performance embedded platforms like the Nvidia IGX Orin and PLC-certified solutions like SICK’s safeVisionary2 enable rapid and accurate safety decisions, even for complex processing models.

3. Wireless Safety Communication
   Advances in high-speed wireless communication, such as those from FORT Robotics, have enabled safer and more responsive robot operations in dynamic environments.  A major gap to be filled here: an open-source safety communication protocol that can widely be used!

4. Human Presence Sensing
   Technologies such as ultrasonic, capacitive, and radar-based sensing (e.g., Inxpect) are now encompassed under IEC 62998. IEC 62998-3, which was just published last year, has started to standardize sensor algorithms and AI applications for safety, a game-changing development for humanoid robot design.

5. Perception for Rule-Based Control
   Rapidly computed perception stacks that output simple binary safety thresholds—like "is a human present?"—are increasingly practical. This ensures compliance with safety requirements while leveraging modern AI and sensor technologies.

Standards and Global Progress for Humanoid Robotics 

Functional safety for humanoid robots benefits from a growing body of standards for general-purpose service and industrial robots, including:

• ISO 13482: Personal Care Robots

• ISO 10218: Industrial Robots

• ANSI/RIA R15.08: Autonomous Mobile Robots

• Regional derivatives like JIS B 8445 in Japan and CSA Z434 in Canada.

Emerging Related Standards and Research

• UL 4600
  A key framework for autonomous product safety that emphasizes system-level hazard assessment. Due to designs like Tesla’s utilization of a vehicle-style CAN bus and embedded hardware, humanoid robots are well set up to be analyzed like vehicles.  Key takeaways from UL 4600, ISO 26262 and ISO 21448 should thus be applied, such as fail-safe mechanism validation, independence of safety systems, continuous monitoring of safety metrics like joint torques, limb trajectories, ground stability, and human proximity, and edge case and uncertainty handling, such as ensuring training over slippery surfaces, uneven terrain, and abrupt human interaction.

• IEC 61508
  Focuses on functional safety of electrical and electronic systems, offering insights for humanoid sensor and control systems.

• BSI 8611
  Provides ethical guidelines for robot safety in public spaces, which is critical for humanoid robots operating outside industrial environments.

Insights from Academia and Policy

Research Contributions

• IEEE RAS Publications: Conferences like ICRA and the Humanoids Conference feature cutting-edge research in humanoid safety.

• TU Delft Robotics Institute: European academic labs contribute to advancements in safety validation and risk assessment methodologies.

• NIST Test Methods: NIST's work on repeatable test methods offers insights into measuring safety performance effectively.

Policy Considerations

• World Economic Forum Reports: Highlight governance and ethical considerations for humanoid robots.

• UNECE Standards: Work on intelligent systems intersects with humanoid robot safety, especially for mobility.

Industry Contributions

Robotics companies and component suppliers are critical to advancing safety:

• Boston Dynamics and Agility Robotics showcase real-world implementations of functional safety in humanoid robots.

• SICK, Velodyne, and Luminar provide cutting-edge safety sensors.

• Tesla participates in global conversations about humanoid robot governance, although adoption of proposals like China's Guidelines for Humanoid Robot Governance remains uncertain in the U.S.

Closing Thoughts

As humanoid robots become more capable and integrated into human environments, functional safety remains the cornerstone of their development. The journey toward robust standards—grounded in innovations, international collaboration, and regulatory rigor—is essential to ensuring that these robots operate safely and reliably. With ongoing efforts from organizations like ASTM, IEEE, and ISO, the future of humanoid robotics looks both promising and secure, with incredible baselines to build off of that we’ve previously covered, such as aerospace and industrial robotics. Contact us at Saphira AI to work with us to make your current products as safe as possible under both existing and upcoming frameworks!

Thank you to Zhaoyuan Gu from Georgia Tech for his assistance and review of this article!