Day 11

For Next Time

• Continue working on particle filter. You should try to have it done by the beginning of Day 12, but the actual deadline is Day 13. You will have a computer vision assignment due on Day 13 as well.

Tuning and Testing Your Particle Filter

In this class you will learn some techniques for tuning and testing your particle filter.

Static Configuration

We'll begin with the simplest form of tuning, static configuration. By configuration we mean setting the values of various constants that control the program's behavior (e.g., number of particles, amount of noise in various updates, etc.). By static we mean that the configuration must be set before launching the node.

• To start, I'll show you how to make the number of particles configurable.
• Then, we'll change likelihood function for the laser scanner.
• Next, I'll show you how to use a YAML file to set a bunch of parameters simultaneously.

Dynamic Reconfigure

Sometimes it is nice to be able to change the configuration of your node dynamically (i.e. while it is running). To do this, we'll need to create a .cfg file to define the configuration of our node.

Tip: to save the configuration parameters to a YAML execute this command.

$ rosrun dynamic_reconfigure dynparam dump /pf new_params.yaml

To load these configuration parameters, run this command.

$ rosrun dynamic_reconfigure dynparam load /pf new_params.yaml

Play around with the built-in particle Filter

If you want to see what this approach looks like taken to the next level, run the built-in particle filter (using the procedure defined in the robot localization assignment). You will see quite a few parameters can be dynamically reconfigured.

Quantitative Benchmarking

Together, we'll write a node to perform quantitative benchmarking of your particle filter.

Parameter Search

Disclaimer: I haven't tried this yet (e.g., to try different values of z_hit). Honestly, it will be a significant amount of work. Here is one possible strategy for doing this.

1. To be consistent in the initialization of your cloud, you will probably want to take in parameters for the initial x, y, theta of your particle cloud in pf.py.
2. You will want to modify benchmarker.py to launch test.launch and play the bag file (e.g., using Popen). Once the bag files has stopped playing (i.e. the process terminates), you will know that the test case is over. When this happens, you can kill the process you used to launch test.launch.
3. You will need to write some sort of loop over the parameter value you want to vary. For each parameter value, perform step 2. During the test you can track the cumulative error in x, y, and theta. After doing this over all the parameter values, you will have a sense of which parameter values are working well and which aren't.