Mean Time Between Interventions

How long, on average, a robot can keep working continuously before a human has to step in.

MTBI measures a robot's reliability over long continuous operation: total operating time divided by the number of human interventions during that time, with a larger number being better. An intervention generally includes a manual emergency stop, a manual reset, a human repositioning an object for the robot, or a human taking over after a task times out. The idea is borrowed from mean time between failures (MTBF) in industrial reliability. The single-trial success rate commonly reported in academic papers is measured after a careful manual reset each time, which hides problems like state gradually drifting or occasional getting stuck during long runs, and also blends speed and accuracy into one number. Deployment-focused teams have therefore started reporting MTBI under a continuous-operation protocol with no resets, measuring it alongside tasks completed per hour to capture both efficiency and reliability.

ExampleThe Habilis-β paper runs a 1-hour continuous-operation evaluation: in a real humanoid-robot logistics workflow, its MTBI is 137.4 seconds, compared with 46.1 seconds for π0.5.

Also called

MTBI

Related

Intervention Rate · Mean Time Between Failures · Success Rate · Real-World Evaluation · Fully Autonomous · Remote Teleoperation Takeover

Sources

Habilis-β: A Fast-Motion and Long-Lasting On-Device Vision-Language-Action Model (arXiv 2602.18813)
Mean time between failures - Wikipedia

As of

2026-02