I create a public class variable called double last_cmd_vel_received that gets initialized in my constructor to last_command_vel_received = 0. Then, in the main() function of my node I read in a ROS parameter called double watchdog_timeout. For the main while loop I use ros::spinOnce() and rate.sleep() (where rate is another ROS parameter value). In my class I have a public function called zeroVelocities() that just sets all the actuator outputs to safe values. So, my main while loop looks something like the following:
...
while (nh.ok())
{
if (ros::WallTime::now().toSec() - my_class->last_cmd_vel_received > watchdog_timeout)
{
my_class->zeroVelocities();
}
ros::spinOnce();
rate.sleep();
}
...
In my callback function for the cmd_vel topic (where the function is part of my class) I set last_cmd_vel_received = ros::WallTime::now().toSec(). You can play with timeout values however you want, but I have found that 0.3 or so works well with my robots.
The general approach is to not add the timeout in the teleop node but in your robot's base node (examples: PR2, Turtlebot). The reason is that it also can happen that your network connection drops. In that case, the robot would still run away.