The site is read-only. Please transition to use Robotics Stack Exchange
| ROS Resources: Documentation | Support | Discussion Forum | Index | Service Status | ros @ Robotics Stack Exchange |
 | 1 | initial version |
Please see EDIT2 listed in my original question. The latency was indeed due to the amount of sensor data being subscribed to by move_base which (I can only presume) shares the same message queue with the /move_base/goal topic. Keeping all four sensor sources in place but dropping the rate from 30 Hz to 15 Hz provides acceptable goal responsiveness from move_base.
| | 2 | No.2 Revision |
Please see EDIT2 listed in my original question. The latency was indeed due to the amount of sensor data being subscribed to by move_base which (I can only presume) shares the same message queue with the /move_base/goal topic. Keeping all four sensor sources in place but dropping the rate from 30 Hz to 15 Hz provides acceptable goal responsiveness from move_base.
EDIT1: Upon further testing, it looks like the latency is due IN PART to the amount of sensor data. However, other factors are involved (e.g., TCP_NODELAY on the /move_base/goal topic, adjusting the queue size, etc.). Once I have this worked out definitively, I'll report back. Sorry for the premature resolution. More testing is required.
| | 3 | No.3 Revision |
Please see EDIT2 "EDIT2" listed in my original question. The latency was indeed due to the amount of sensor data being subscribed to by move_base which (I can only presume) shares the same message queue with the /move_base/goal topic. Keeping all four sensor sources in place but dropping the rate from 30 Hz to 15 Hz provides acceptable goal responsiveness from move_base.
EDIT1: EDIT1: Upon further testing, it looks like the latency is due IN PART to the amount of sensor data. However, other factors are involved (e.g., TCP_NODELAY on the /move_base/goal topic, adjusting the queue size, etc.). Once I have this worked out definitively, I'll report back. Sorry for the premature resolution. More testing is required.
EDIT2: I believe I have this issue resolved. On the advice of @demmeln a combination of throttling my sensor sources and using a multi-threaded spinner in move_base has provided acceptable latency for the system. The specifics for my setup that work well are to scale the frequency of my 4 sensor sources down from 30 Hz to 15 Hz and to use a multi-threaded spinner within move_base with 2 threads. I have submitted a pull-request for the move_base changes. Those changes are minimal and available here: https://github.com/ros-planning/navigation/pull/182
I should note that all changes that I had originally made to actionlib (i.e., TCP_NODELAY and changing the queue_size) were reverted during my testing today. The changes that worked for me were limited to what I describe in the preceding paragraph.
Thanks for the help / input.
ROS Answers is licensed under Creative Commons Attribution 3.0 Content on this site is licensed under a Creative Commons Attribution Share Alike 3.0 license.
