ROS Resources: Documentation | Support | Discussion Forum | Index | Service Status | ros @ Robotics Stack Exchange
Ask Your Question

How a robot navigation algorithm can be evaluate? [closed]

asked 2012-07-09 04:12:44 -0500

abdullah gravatar image

updated 2012-07-12 07:53:24 -0500


I have implemented a potential field algorithm for robot navigation? How can I test this technique. Is there any best way to evaluate this algorithm?

What I am doing now is that I have set an environment having obstacles in gazebo. And when a goal is given to the robot, it reaches at goal position avoiding all obstacles. But when I am repeating same experiment with same environment, robot has some different trajectory every time.


After 2 days I have run some experiments in stage simulator. Again robot have not same trajectory every time. I have three scenarios here.

  1. Run a robot without any obstacle. Without obstacles robot have same trajectory every time I repeat the experiment. It shows robot movement is very accurate.
  2. Run a robot with 1 obstacle in the way to the goal. Every time robot has same trajectory, but there was a little difference. Or we can say trajectory was 95% same.

  3. Run 2 robots, moving in opposite direction. On repeating this experiment every time I have some different trajectories.

Please Can anyone tell me that, Why it has this kind of behavior? Is there any noise in sensor data of stage simulator? I am running all my nodes in a launch file. I am considering that all my nodes start working at the same time. This could be effecting my experiments.


edit retag flag offensive reopen merge delete

Closed for the following reason the question is answered, right answer was accepted by Procópio
close date 2016-10-03 08:08:30.461348

3 Answers

Sort by » oldest newest most voted

answered 2012-07-09 04:39:28 -0500

Gazebo simulation often is not 100% deterministic as minor timing variations of the real-time dynamics simulation leads to small (sometimes also large :) ) jitter. We get pretty consistent results when simulating/evaluating path-planning/SLAM and other high level software however. Sounds like your potential field implementation might not be robust against such jitter with the parameters you used. Have you tried it with another simulator or real robot?

edit flag offensive delete link more


Thanks, Yes my potential field implementation is not so robust. And I have not tried it with any other simulator or real robot. But I think that I will try it in simulator stage. What do you think that how can I make my algorithm robust?

abdullah gravatar image abdullah  ( 2012-07-09 04:52:57 -0500 )edit

For an answer, this could be a separate question (with lots of additional information about what you are doing).

joq gravatar image joq  ( 2012-07-09 05:47:00 -0500 )edit

OK, Please can you suggest me some basic techniques of testing navigation algorithms. Or Is there any general rules to test robot navigation? Thanks

abdullah gravatar image abdullah  ( 2012-07-09 06:05:06 -0500 )edit

answered 2012-11-29 07:26:09 -0500

abdullah gravatar image

For testing and evaluating robot navigation

I was able to evaluate navigation algorithm using 4 testing parameters.

(1) total time (in seconds) to complete navigation task

(2) path length (in meters)

(3) curvature change

(4) Lateral stress.

I got help from an article "Performance evaluation of pure-motion tasks for mobile robots with respect to world models" by Daniele Calisi and Daniele Nardi

edit flag offensive delete link more


hi, @abdullah, could you say how did you measure path length traveled by each robot? (there is an option as described here, but is there some better way for stage simulations?

gavran gravatar image gavran  ( 2016-09-30 12:21:42 -0500 )edit

hi, I had used the same method that you have mentioned.

abdullah gravatar image abdullah  ( 2016-10-30 09:01:09 -0500 )edit

answered 2012-07-11 12:32:25 -0500

KruseT gravatar image

Define a measure, then run many simulations, and evaluate the performance statistically. E.g. measure the time the robot required to get to the goal, the distance travelled, how many times the robots collided, how close the robots came to each other, things like that.

edit flag offensive delete link more


Thanks for your reply. Yes I will do the same. But as I mentioned in my question that repeating experiment every time I have some different trajectory.

abdullah gravatar image abdullah  ( 2012-07-12 09:26:00 -0500 )edit

can you share me the work that you have been doing simulations with potential field

alexi gravatar image alexi  ( 2015-10-20 16:06:05 -0500 )edit

Question Tools



Asked: 2012-07-09 04:12:44 -0500

Seen: 1,515 times

Last updated: Nov 29 '12