Built to operate safely around people.
Yantra builds machines that act in the physical world, alongside people. Safety is a design constraint from the first line — in how the body moves, in the human authority over it, in how capability is tested, and in how a system reaches the field. This work is in active development, and we describe where it is heading rather than claiming a finished record.
| Compliant | force-limited |
| Human | in the loop |
| Fail-safe | by default |
| On-device | and logged |
A person can always supervise, intervene, and stop the machine. Safety is the constraint we design around, not bolt on.
Five commitments, by design — how we build for safety, stated as constraints on the work rather than claims about a finished record.
Physical safety
A machine that shares space with people must be safe in the way it moves, not only in what it decides.
| 01 | Compliant control — the body yields on contact rather than driving through it. |
| 02 | Force and speed are limited to bounds set for the task and the surroundings. |
| 03 | Fail-safe stops bring the system to a safe, stable state on fault, loss of power, or loss of confidence. |
| 04 | Whole-body coordination keeps motion legible, so a person nearby can read intent. |
Human oversight
A person stays in charge. The system is built so a human can always supervise, intervene, and stop it.
| 01 | An accessible stop that takes precedence over autonomous behaviour, at all times. |
| 02 | Human-in-the-loop approval for consequential actions, rather than unattended commitment. |
| 03 | Clear indication of what the system is doing and what it intends to do next. |
| 04 | Authority that defaults to the operator — autonomy is granted within bounds, never assumed. |
Reliability & testing
Capability earns its way into the world through staged validation, not a single leap to deployment.
| 01 | Simulation first — behaviour is exercised and stressed before any hardware moves. |
| 02 | Then supervised real-world operation, with a person present and limits in place. |
| 03 | Continuous monitoring of the system's own state and confidence during operation. |
| 04 | Scope widens only as evidence accumulates — capability by capability, not all at once. |
Secure & auditable
Control lives on the robot, and what it does is recorded. There is no quiet hand on the system from elsewhere.
| 01 | On-device control — the decision loop runs on the robot, not on a remote server. |
| 02 | Actions are logged, so behaviour can be reviewed and accounted for after the fact. |
| 03 | No silent remote control — the system is not steered invisibly over a network. |
| 04 | Designed to be inspected: a synthesised controller is a defined model, not an opaque box. |
Deployment safeguards
How a system reaches the field matters as much as how it is built. Rollout is deliberate and supported.
| 01 | Gated rollout — deployment expands in measured stages, not by default everywhere. |
| 02 | Operator training before responsibility for a system is handed over. |
| 03 | Operating limits and procedures defined for each setting the system enters. |
| 04 | A path to report, pause, and roll back when something doesn't look right. |
Honest about the work ahead.
These are the commitments we build to, not a claim that the job is finished. Yantra's systems are in active development; the safety work matures alongside the capability it protects, and we will not present it as more proven than it is.
Building machines that operate safely around people is hard, and earning that trust takes staged evidence over time. We would rather state plainly where the work stands than overstate it — and let the record, as it accumulates, speak for itself.
Safety and human oversight are designed in from the first line, not added at the end.
Our responsibilities
How we hold ourselves accountable.
Safety is one part of a broader commitment to building embodied intelligence responsibly — with human oversight, and toward AGI pursued with care.