📩 Aperture@thegaogroup.com
Verified AI Decisions and Physical Control
Physical AI and AIoT Research Team of
Aperture Venture Studio, GAO RFID Inc. and GAO Tek Inc.
This program connects the AI Decision Engine to the Physical AI Action Engine. Its central concern is turning recommendations into controlled, measurable physical outcomes.
Verified AI Agents for Industrial Device and Equipment Control
Develop agents that translate an operational objective into a sequence of authorized actions through industrial gateways, APIs or controller interfaces.
The agent should check equipment state and action constraints, obtain approval when required, execute through a controlled interface and confirm the result. Initial work should use existing models and bounded action sets.
Research questions
Can the agent select valid actions from ambiguous requests? Can it detect that a command succeeded electronically but failed physically?
Initial demonstration
An agent operates a pump, fan or conveyor to achieve a defined objective.
Evaluation
Task success, invalid-action attempts, constraint violations, outcome-verification accuracy and recovery performance.
Constraint-Aware Adaptive Control and Industrial Optimization
Study how AI can improve equipment operation while respecting limits on temperature, pressure, speed, energy, quality and equipment wear.
Begin with supervisory decisions, such as choosing permitted setpoints or operating schedules. More demanding control tasks can follow where the team has the required process knowledge and test facilities.
Research questions
Can learned models improve control under changing conditions? How should uncertainty limit the actions a controller may choose?
Initial demonstration
Optimize a pump or thermal system while maintaining specified operating bounds.
Evaluation
Energy or throughput improvement, tracking error, constraint violations and performance under disturbances.
AIoT and Physical AI for Building and Energy-System Operation
Develop methods to determine whether incoming measurements are trustworthy before they influence an AI decision. Address sensor drift, calibration errors, communication gaps, incorrect units, timestamp misalignment and device substitution.
This deserves its own topic because a system must distinguish a failing machine from a failing sensor. It also provides a reusable capability across Aperture’s portfolio.
Research questions
How should occupancy, weather and equipment state affect operating decisions? Can the system balance energy consumption with comfort or process requirements?
Initial demonstration
Control a ventilation or cooling testbed, or run a supervised pilot with a facilities partner.
Evaluation
Energy consumption relative to a normalized baseline, comfort or process compliance, peak demand and operator interventions.