Developing a complete robotic system often requires combining multiple behaviours into a complex decision grid, with elements running in sequence or in parallel, eventually interrupting each others.
To solve this “age-old” problem, ROS provides two main tools:
Actionlib: a client-server architecture that provides a way to specify results to be achieved. While the server works on these results, it should report progresses and ultimately report when the task is completed.
Smach: a python API to define complex state machines. It can interact with ROS services and actions to define a complex behaviour, including nesting, interruptions and concurrence.
Combining Smach and Actionlib, one could build arbitrarily complex systems. Hence, why would another task management system be necessary?
The main argument in favour of our task scheduler is the simplicity of its use, particularly in comparison with Actionlib. As usual, simplicity is a trade-off against expressiveness. Simplicity can be sacrificed by linking our task scheduler with Actionlib and/or Smach to exploit the best of both worlds.
Our task manager is presented in a small article on HAL, and the framework is available on github. The video below illustrates what it can do in the context of a UAV-UGV collaboration inspired by the needs from the Flourish project (thanks to Badr El Hafidi for preparing the video).
Our task manager is also used in the following industrial demonstrator: