For decades, we designed robots not to read the room.

We removed variability.
We removed ambiguity.
We removed humans from the equation wherever possible.

That’s how we achieved safety.

Now we are doing the opposite.

From Controlled Systems to Human Environments

As humanoid robots move beyond cages and into shared spaces, the requirement is no longer just task execution—it is interpretation.

These systems must:

• Recognize people and movement
• Understand dynamic surroundings
• React in real time
• Communicate intent clearly

In short, they must operate in the same unpredictable environments humans navigate instinctively 

This is not an incremental change.
It is a fundamental shift in what we expect robots to do.

When Behaviour Can’t Be Fully Defined

Traditional functional safety is built on a simple premise:

If we can define behaviour, we can verify it.

But once a robot starts “reading the room,” behaviour is no longer fixed.

Actions depend on:

• What the system perceives
• How it interprets context
• How quickly it decides and responds

A safe outcome is no longer just about force limits or stopping distances.
It depends on whether the robot correctly understands the situation.

And that introduces something functional safety has always tried to minimize: uncertainty.

From Mechanical Hazards to Situational Risk

In traditional robotics, hazards are mostly physical:

• Unexpected movement
• Excess force
• Loss of control

In context-aware systems, new hazards emerge:

• Misinterpretation of human intent
• Delayed or inappropriate response
• Behaviour that is technically correct—but contextually wrong

The robot may be operating “within specification”… and still create risk.

Safety is no longer just engineered into components. It becomes a property of the interaction between robot, human, and environment.

Why This Breaks Traditional Certification Thinking

Standards such as ISO 10218—and even their evolution toward application-based safety—are still rooted in the idea that risks can be identified and addressed upfront.

But with adaptive, perception-driven systems:

• Not all scenarios can be anticipated
• Not all behaviours can be fully tested
• Not all risks can be eliminated at design stage

Instead of asking:

“Is the robot safe?”

We are forced to ask:

“How does the system remain safe as conditions change?”

That’s a fundamentally different problem.

Toward Continuous, Contextual Assurance

If behaviour is dynamic, safety must be too.

This shifts assurance from:

• One-time validation → continuous validation
• Fixed limits → context-dependent performance
• Component focus → system-level understanding

It also introduces new requirements:

• Confidence in perception systems
• Transparency of decision-making
• Clear communication of intent to humans
• Ongoing monitoring of real-world behaviour

Because in human environments, safety is not just about avoiding harm—it is about maintaining trust.

Conclusion

We are moving from robots that operate in controlled environments…
to robots that must operate in human ones.

That means moving from:

• Deterministic safety
  to
• Contextual safety

And that changes the role of functional safety and certification.

It is no longer enough to define safe behaviour at design stage.
We must ensure the system can continue to behave safely—moment by moment.

Because once robots start “reading the room,”
safety is no longer something you certify once.

It is something the system must continuously achieve.