ROS path planners such as MoveIt! and Descartes are good at finding feasible robot paths that avoid obstacles and optimize motion. In some applications, adjustments to nominal paths must be made based on sensor feedback or operator input. The issues were raised at ROSCON 2015 . The specific application driving this need is robotic assembly using end-effector force feedback, where parts such as computer board components need placement in holes or sockets with tight clearances, and the object locations may vary considerably with a need to locate them using vision systems. Achieving reactive and adaptive path planning may require extensions to several parts of ROS and ROS-Industrial, including the MoveIt! or Descartes path planners, the ROS control layer, or the Simple Message protocol to hardware robot controllers. Development can take place without actual force sensors or cameras, using simulated sources of the high-level outputs of these sensors in terms of small adjustments that affect the trajectories. The key software objective is to put in place a method to adjust trajectories after they have been nominally planned, rather than focusing on sensor processing.


Kentaro Wada


  • Fred Proctor