Safety
Safety before motion
Nothing physical happens until simulation and a human sign-off say it may. That is the product, not a disclaimer.
Your safety rules and facility layout stay yours. Volant only ever proposes actions inside them.
The system adapts to reality in real time. It never adapts its own permissions.
Simulation-first deployment
Every approved graph is exercised for collision, timing, and limit violations in simulation against your facility model. Failures stay in software.
Human-in-the-loop approval
Operators see a preview and a diff. Execution is a deliberate act, not an autonomous leap from a chat prompt to a moving robot.
Bounded real-time adaptation
Setup is guided and visual: zones, hardware limits, and approval thresholds are configured on a facility map and with preset sliders — not by ambiguous free-form language at the permissions layer. Day-to-day tasks are then given in plain language inside that envelope. During execution the system may reroute, re-sequence, or pause without a new approval, because those moves stay inside what was already signed off. Expanding permissions always returns to simulation and a human gate. Designed to align with the EU AI Act and the Machinery Regulation — not certified until certification is achieved.
Designed to align with EU law
We describe Volant in the context of high-risk AI system duties under the EU AI Act and of machinery safety under the EU Machinery Regulation. Language on this site is “designed to align with” — not “certified.”