Independent safety runtime
for real-world robots.
We help robotics teams test and deploy robots in dynamic environments by detecting risk, triggering interventions, and logging safety events for evaluation.
For teams testing robots beyond controlled demos.
Problem
Robots are leaving controlled environments.
As robots move from controlled cells to dynamic human environments, teams need more than manual supervision, e-stops, static rules, or vendor-specific safety logic. They need runtime visibility into when, why, and how risk appears.
Human-in-the-loop safety
Teams often rely on someone watching the robot, ready to intervene manually.
Stack-specific safety
Safety logic is often tied to a specific controller, robot stack, or manufacturer workflow.
Limited incident visibility
When something unsafe happens, teams often lack structured logs, replay, and evaluation data.
Solution
A policy-independent safety layer.
Our runtime sits outside the robot policy. It observes the scene, estimates risk, intervenes through the robot control stack, and records what happened.
- Runs as a separate process, outside the robot policy.
- Intervenes through the existing control stack — pause, slow, or block.
- Records every intervention for replay and evaluation.
Camera / sensors
RGB · depth · workspace views
Scene understanding
humans · objects · robot state
Risk monitor
continuous risk estimation
Intervention layer
gates commands to the robot
Robot control stack
your policy · any vendor
Logs · replay · evals
structured safety events
Demo
From risk detection to safe resume.
Three moments from a single test episode with our prototype runtime.
Human reaches into workspace
Hazard detected and robot paused
Scene clears and robot resumes
Who it's for
For teams testing and deploying robots in the real world.
Robot learning teams
Evaluate learned policies on physical robots with an external safety monitor.
Robotics startups
Capture unsafe events and debug risky behavior during hardware experiments.
Industrial robotics teams
Monitor robot workspaces and add visibility beyond static safety rules.
LeRobot / ROS teams
Connect a safety layer to open-source and custom robot control stacks.
Observability
Interventions should be observable.
A safety runtime should not only pause the robot. It should explain when, why, and how it intervened.
| Time | Trigger | Risk | Action | Status |
|---|---|---|---|---|
| 00:12.401 | hand near sharp object | high | pause | resolved |
| 00:18.932 | workspace cleared | low | resume | safe |
| 00:24.110 | object moved unexpectedly | medium | slow | monitoring |
Conversations
How do you handle robot safety today?
We're talking with robotics teams, labs, startups, and industrial operators about how they monitor risk, handle interventions, and analyze safety events during real-world robot testing.