Summary of Key Points
Robot safety has evolved from a mere technical issue to a systematic endeavor involving international governance, regional regulation, corporate practices, and academic research. As robots move beyond industrial enclosures into human-robot collaboration scenarios (such as humanoid and mobile robots), the security risks associated with physical AI have shifted from being theoretical errors on screens to real-world collisions or malfunctions. Various stakeholders are working together to establish a clear chain of responsibility that ensures robots go from being capable of performing tasks to being deemed safe for use.
Detailed Analysis
1. International Governance: Robot Safety Is No Longer Just Corporate Business
In the past, robot safety was primarily the responsibility of individual companies developing these technologies. However, it has now become part of the broader global AI governance framework. For example, at this year’s AI for Good summit in Geneva, Switzerland, United Nations Secretary-General António Guterres highlighted three critical issues: How can robots be tested to ensure they are safe? How should risks be assessed? And who is responsible in the event of an accident? The role of international governance is not to dictate specific sensor requirements for robots but to set clear boundaries—identifying capabilities that cannot be commercially used without proper safety measures, specifying which incidents must be reported, and requiring third-party evaluations. This places long-term pressure on companies, as ensuring robot safety is no longer a matter solely for the R&D department; it has become an essential capability that needs to be demonstrated to customers, investors, and regulatory authorities, just like data security or carbon emissions.
2. Regional Regulation: EU Regulations Make Safety a Prerequisite for Market Entry
The truly challenging aspect for companies comes from regional regulations. For instance, the new European Machinery Directive, which will come into effect in 2027, applies to the next generation of robots equipped with AI. To enter the European market, companies must meet these requirements:
- The CE mark cannot be obtained simply by purchasing a product; companies must conduct their own risk assessments, prepare technical documentation, and prove that their products comply with the regulations.
- Compliance with standards (such as ISO 10218 for industrial robots) does not guarantee safety. The actual risks can arise from specific components (like metal-cutting tools) or the environment (such as crowded storage areas). In the future, robots that cannot meet safety requirements may not even get the opportunity to undergo pilot tests—safety will become a bargaining chip alongside price and performance.
3. Corporate Practices: Shifting from Physical Barriers to Systematic Risk Management
Industrial robots used to rely on physical barriers and emergency stop buttons for safety, but mobile and humanoid robots need to operate in crowded environments where such barriers are ineffective. Companies are adopting new approaches:
- Separating the “task layer” (where the robot performs tasks) from the “safety layer” (which monitors and controls potential risks). For example, the safety system can prevent the robot from moving if it cannot see its surroundings clearly or if signals are unreliable.
- The environment itself also plays a role in ensuring safety. For instance, when an autonomous forklift enters a trailer, onboard sensors may have blind spots, so external cameras and environmental systems provide feedback to guide the robot (a “from-the-outside-in” approach).
- Companies must provide customers with detailed information about the robot’s safety limitations, such as its maximum operating speed or the need for re-assessment when changing loads.
4. Academic Research: Filling the Gap in Human-Robot Interaction Testing
Current robot testing often takes place in controlled environments with fixed objects and smooth surfaces. However, in real-world scenarios, people may touch, push, or suddenly release robots. Academia is working to address this gap by collecting data on human-robot interactions (e.g., pushing and pulling during collaborative tasks) and using simulations to study how robots respond to such forces. For example, some robots that appear stable in normal operations may become unsteady when pushed, potentially causing injuries. Such research reminds the industry that robots must not only be capable of performing tasks but also able to remain stable under unexpected conditions.
5. A Closed Loop of Responsibility
Ensuring robot safety is a collective effort involving all stakeholders:
- Robot manufacturers must provide accurate information about the robot’s maximum load, speed, and how it will behave in case of failures (e.g., whether it will squat down instead of falling).
- Integrators need to assess the combined risks of the robot, its tools, and the surrounding environment.
- Site operators should not make changes to the workplace (such as adding shelves) without considering safety implications.
- Software providers must manage software updates to ensure that new versions do not bypass existing safety measures.
Only by closing this entire chain of responsibility can robots transition from being experimental prototypes to reliable components in production environments.
In Conclusion
For robots to work effectively alongside humans, it is essential to have a comprehensive safety framework that encompasses international regulations, corporate best practices, rigorous academic testing, and clear responsibilities for all stakeholders. After all, no one wants to collaborate with an “out-of-control mechanical entity.”