Subvertio AI: a brain for robots

Subvertio AI is developing the computing hardware that helps a robot interpret sensor information and decide how to respond. Its proposed architecture combines rapid reflexes with more detailed processing, like reacting to unexpected contact while working out the next movement.

Subvertio is a separate related company developing specialized AI inference processors. Its planned compact robotics computer is distinct from its liquid-cooled workstation product. Manufactured silicon and robot integration remain development milestones.

 

Patent pending, in plain English

Specialist chips working together. Small chips share sensing, conventional computing and three-state logic. Three-state logic uses negative, neutral and positive values to simplify suitable AI calculations.

A shortcut from sensing to response. A proposed fast path turns sensor signals directly into three states. A second path performs more detailed analysis and can refine or override the initial response.

U.S. provisional application 64/142,758 (FAV_012), submitted August 27, 2026. The supplied design includes simulated and placed-and-routed circuitry; the proposed reflex speeds and complete system are not measured product results. Explore Subvertio AI.

 

Favis Advanced Robotics

FAR is developing Seraphim, a humanoid robot for industrial facilities, maintenance and critical infrastructure. Proposed tasks include moving supplies, assisting technicians and inspecting equipment in spaces built for people.

The development program brings together a human-like body, a proposed Subvertio onboard computer, multispectral eyes and learning from human demonstrations. Current materials describe simulation work; physical capability and customer trials are ahead.

 

More ways to sense and learn

Multispectral eyes (FAV_013, patent pending): combine ordinary images with infrared and thermal views to help identify conditions such as unusual heat.

Human demonstration (FAV_014, patent pending): a proposed suit and eyewear capture movement, muscle activity and contact pressure to help teach robots how a person performs a task.

These are proposed technologies described in provisional applications, not granted patents or demonstrated product features. Explore Favis Advanced Robotics.

 

Talk with Steve

Discuss a power project, AI computing requirement, robotics application or partnership. Leave your number above; name, company and email are optional. Steve will help identify the relevant technical information and next step.