Optimization example

Download: example.tar.gz

Description

This example uses the optimization framework to optimize a specific behavior of a robot (the e-puck) in webots. It is a very simple example which should be easy to understand and explain the general principles of using the framework.

The behavior optimized in this example is to have the e-puck drive a single, full circle in a specific amount of time. This can of course be solved analytically very easily, but it also serves as a nice example for optimization.

The fitness function used to optimize on this behavior contains two separate objectives.

  1. Maximize the traveled distance (or radius)
  2. Minimize the distance to the original starting position of the robot after N seconds

One way to combine these two objectives in a single fitness value can be:

Maximum distance - End distance to origin

This will nicely converge on a solution that maximizes the maximum distance, and minimizes the end distance to the origin.

Running the example

To run the example locally on your computer, you can follow the following steps:

  1. Modify jobs/example.xml to set the absolute path to the webots world on your computer (webots/worlds/example.wbt)
  2. Compile the two webots controllers in webots/controllers/
  3. Start one optimaster
  4. Start at least one optiworker
  5. Run optirunner jobs/example.xml

Steps 1 and 2 only have to be done once.

After the job is finished, you can use optiextractor to open the results database (jobs/example.db.1) and inspect the results. If you run the optimization in webots batch mode (the default), you can use the ‘override’ section of the optiextractor to set the webots mode to “run” so that you can actually see the result. If you than replay the best solution you should see the e-puck robot making one full circle in the amount of time specified in the job.

Changing optimization parameters

You can play with the parameters in the job description. For example, you can see what happens when you change the number of particles, or iterations of the PSO. You can also easily change the maximum allowed speed for each wheel, or change the fitness function in the job file. Then rerun the job and inspect the results.

Package contents