TrustRobotics Releases a Working Draft of the TrustBoundary Humanoid & Physical AI Safety Standard (TB-PAS 1.0)
Updated: Aug 16
For decades, robotics safety has focused on machines: emergency stops, speed limits, force thresholds, protective separation, and functional safety.
The age of Humanoid & Physical AI introduces a new question:
When should an AI-generated action be permitted to move the physical world?
TB-PAS 1.0 proposes that every AI-generated physical action pass through a mandatory TrustBoundary before reaching a robot's actuators.
The core principle is straightforward:
A conforming Humanoid & Physical AI system shall prevent an AI-generated candidate physical action from reaching an actuator-control endpoint unless a TrustBoundary has produced a valid release authorization for that candidate physical action.
The standard defines a runtime safety pipeline:
Generate → Intercept → Normalize → Predict → Evaluate → Authorize → Enforce → Monitor → Revoke or Complete
Rather than prescribing a particular AI model, robot operating system, processor, or hardware architecture, TB-HPAS defines the observable safety behaviors every Humanoid & Physical AI system should provide.
The initial working draft introduces:
• A common technical vocabulary for Humanoid & Physical AI safety
• Progressive validation classes, from deterministic constraints through predictive world-model validation
• A humanoid-specific safety profile addressing whole-body stability, fall consequences, manipulation, human interaction, and public-space operation
• A Physical-Authority Token that binds authorization to a specific robot, action, context, and execution envelope
• A certification framework with repeatable conformance tests and measurable pass/fail criteria
Existing robotics standards answer important questions about how robots should be designed and integrated safely.
TB-HPAS focuses on a different layer:
How AI-generated physical actions are evaluated, authorized, monitored, and, when necessary, prevented from executing.
As Humanoid & Physical AI systems transition from research labs into factories, hospitals, homes, schools, warehouses, public infrastructure, and transportation systems, runtime governance will become as important as model performance.
TB-HPAS 1.0 is being released as a working draft to encourage technical discussion among robotics manufacturers, AI developers, safety engineers, researchers, insurers, standards organizations, regulators, and the broader Humanoid & Physical AI community.
If we can establish common interfaces and measurable runtime safety requirements today, we have an opportunity to improve interoperability, certification, and public trust as the next generation of intelligent machines enters the physical world.
We welcome constructive technical feedback from the robotics and AI community.
Comments