TrustBoundary™

Version: 0.1 Draft

TrustBoundary™ is TrustRobotics' execution-governance architecture positioned between AI-generated robot intent and physical actuation.

What This Is

Rather than allowing an AI model, VLA, planner, or application to directly control actuators, TrustBoundary receives a Candidate Action and determines whether that proposed physical action should proceed.

The canonical flow is:

Vision / Lidar / Speech / Force / Joint State
                    |
                    v
              VLA / Planner
                    |
                    v
             Candidate Action
                    |
                    v
+------------------------------------------+
|               TrustBoundary              |
|                                          |
|  Predict Future Physical Consequences    |
|  Validate                                |
|  Apply Policies                          |
|  Approve / Modify / Reject               |
+-------------------+----------------------+
                    |
                    v
          Robot Controller / Runtime
                    |
                    v
                 Actuators

Repository Structure

TrustBoundary/
├── README.md
├── spec/
│   └── vla-action-command/
│       └── README.md
├── reference/
│   └── vla-action-gate/
│       └── README.md
└── future components
    ├── validators/
    ├── policy/
    ├── execution-envelopes/
    ├── actuator-firewall/
    └── sdk/

The VLA Action Command specification and VLA Action Gate reference are now maintained as parts of TrustBoundary rather than as separate TrustRobotics products.

Core Components

TrustBoundary may include:

• Candidate Action objects;
• forward or predictive physical-state evaluation;
• validator selection and orchestration;
• policy evaluation;
• validator graphs;
• execution envelopes;
• release authorization;
• actuator-firewall interfaces;
• audit logging;
• approve / modify / reject decisions.

Why It Exists

Traditional robot software is generally optimized to answer questions such as:

• How can the robot perform this action?
• What trajectory should it execute?
• Can the controller satisfy the requested motion?

TrustBoundary adds another question:

Should this physical action be permitted to execute?

The objective is to create a deterministic trust layer between probabilistic AI decision-making and physical action.

Relationship to the TrustRobotics Stack

TrustIntelligence™
OBSERVE / QUERY / UNDERSTAND
          |
          v
TrustRouter™
DETECT / GOVERN / COORDINATE
          |
          v
TrustBoundary™
PREDICT / VALIDATE / AUTHORIZE
          |
          v
Robot Runtime / Controller
          |
          v
ACT

Observe → Govern → Validate → Act

TrustRouter provides premises-level robot identity, network governance, shared environmental services, and heterogeneous robot coordination. TrustBoundary governs whether candidate physical actions proceed toward execution.

Relationship to the Humanoid Constitution

The Humanoid Constitution defines machine-enforceable rules, authority, priorities, and governance domains for humanoid operation.

TrustBoundary is an execution point at which applicable rules, validators, state, and authority may be evaluated against a proposed physical action.

The two concepts therefore complement rather than replace each other:

Humanoid Constitution
rules / authority / priorities
          |
          v
TrustBoundary
runtime candidate-action evaluation
          |
          v
Actuator Runtime

Integrated VLA Interfaces

• spec/vla-action-command/ — proposed action-command interface between VLA/planner output and TrustBoundary.
• reference/vla-action-gate/ — minimal TrustBoundary-compatible gate used for prototyping and developer education.

The former standalone repositories remain only as compatibility pointers for existing links.

Related TrustRobotics Repositories

• TrustRobotics/TrustRouter — Robot Network Governance and Robot-Enabled Access Point architecture.
• TrustRobotics/humanoid-constitution — machine-enforceable humanoid governance rules and authority.
• TrustRobotics/humanoid-os — humanoid-specific operating-system architecture.
• TrustRobotics/TrustRobotics-rfcs — public RFC series.
• TrustRobotics/TrustRobotics-architecture — umbrella architecture.

Design Objectives

• vendor-agnostic Physical AI governance;
• AI-model independence;
• form-factor independence;
• physical safety support;
• deterministic execution authorization;
• explainable governance decisions;
• auditable AI-to-actuator trust;
• runtime validation before physical execution.

Status / Implementation State

Draft architecture and reference-interface stage. Public repositories may contain limited reference components, but do not represent a complete safety-certified TrustBoundary product.

Patent and Licensing Notice

TrustBoundary technologies are associated with issued and/or pending TrustRobotics patent rights. Publication of this repository does not grant a patent license. Commercial implementation of protected TrustBoundary technologies may require a separate TrustRobotics license.