Embodied AI Glossary中文

Watchdog

看门狗Advanced

A timer that requires a program to check in periodically and forces a safe state if it stops responding in time.

A watchdog was originally a hardware or software timer in embedded systems: while a program runs normally, it must periodically ‘feed the dog’ (kick it, resetting the timer); if the program hangs or crashes and misses a feeding, the timer expires and triggers a corrective action, usually putting the output into a safe state (motors off, high voltage cut) before restarting. In robotics its most common use is as communication-timeout protection: the host computer or policy process must send commands at no less than some minimum frequency, and a timeout triggers braking, a damping mode, or a protective stop — preventing the robot from continuing on its last command if the network drops or the process crashes. For example, the ur_rtde library provides setWatchdog and kickWatchdog interfaces that, by default, require updates of at least 10 Hz or shut down control; ROS 2's differential-drive controller, diff_drive_controller, has a cmd_vel_timeout parameter that by default treats a velocity command as stale if no new one arrives within 0.5 seconds. Deploying a learned policy on a real robot generally requires a watchdog configured at this low level.

ExampleA mobile base is teleoperated over Wi-Fi from a laptop, and the signal suddenly drops. The base controller hasn't received a new cmd_vel in 0.5 seconds, so the old command is judged stale and the base stops — rather than continuing to drive forward at its last commanded speed.

Also called
Watchdog Timer, WDT, Communication Timeout Protection
Related
Emergency Stop · Protective Stop · Damping Mode · Real-Time Control · Control Latency · Functional Safety
Sources
Wikipedia: Watchdog timer
ur_rtde: rtde_control_interface.h(setWatchdog / kickWatchdog)
ros2_controllers: diff_drive_controller(cmd_vel_timeout)

See it in the full glossary →