Running

[ghost]

Hi,

I think I have set everything up correctly, where should I run 'director simulator.py' in order to properly produce a result? is their a specific command that should be used? Thanks!

[anassinator]

It should be from the root of this repository, but it is possible this wont run anymore because director and tensorflow APIs were unstable them and have changed a lot since. I'm currently rewriting it to remove the director dependency and use a stable version of tensorflow to help with that.

[ghost]

That might explain... oh great would definitely like to keep updated on that! Will sling you an email @hello@anassinator.com

[ghost]

can i ask when you might be finished with your up to date version? look forward to seeing it

[anassinator]

Sorry I’m on vacation this week. I’m hoping for sometime in the next two weeks.

[ghost]

awesome, enjoy!

[ghost]

Any update on the progress on removing the director dependancy?

[anassinator]

Sorry I've been distracted by some other tasks. So far I have managed to completely replace the director dependency with a simple pygame simulation, but still need to get the rest updated and interfacing with it.

[ghost]

That sounds great, pygame has proved really useful. Any idea when you might complete it? Would love to give it a try.

[ghost]

Managed to get it running by using ubuntu rather than Mac, awesome stuff! Would love to learn more about the theory behind this, have you created any written work I could read on this topic?

[anassinator]

Sorry I took so long to get back to you. I've been busy so I had put this on hold for a while, but I will pick it back up this week. I can't really commit to an ETA to be honest.

I haven't written any paper explicitly on this, but I have outlined this in one section of a larger project. I'm attaching the relevant parts below.

dqn-obstacle-avoidance.pdf

[ghost]

No problem at all. I managed to get the simulator up and running on Ubuntu and I'm amazed by the result, awesome work!

I have 2 quick questions after reading your document. Where might I be able to change the Hyper-parameters such as Learning rate, discount factor within the code if possible? How might I zoom in on the simulator to get a better view?

Thanks!

[anassinator]

When you run director simulator.py --help you will find those options available as arguments to tweak.

As for zooming in, I don't remember and I don't have director set up right now. I believe it was just a mouse wheel or something. Otherwise, you might be ale to find something in director's user guide.

[ghost]

thanks for the help!

[ghost]

One more question sorry!

running 'director simulator.py --help' pulls up the option for arguments for obstacle density, learning rate etc. Apologies if it's obvious but how would I change these values and run?

This work has been very helpful, would love to be able to donate if you have somewhere I can do that!

[anassinator]

No worries! You can add command-line arguments as follows:

director simulator.py --learning-rate 0.1 --exploration 0

The order of the arguments doesn't matter. You would want to set the learning rate and exploration to 0 when done training to get nice results. Ideally, this would simply decrease automatically, but I haven't had time to do that yet.

Thanks for the kind words! I don't have anything set up for personal donations, but this work is being used for a drone we're developing at McGill Robotics (a non-profit student organization) and we will be accepting donations in our upcoming Seeds of Change campaign next week if you're interested. Thanks again!

[ghost]

Great that works well thanks!

I'd be more than happy to make a donation to the campaign, just send me a link! Thanks

[ghost]

Hey! If the opportunity is there I would still like to donate to the campaign as thanks for the help provided!

[anassinator]

Hey @jman123456

Sorry, the campaign was delayed for a bit, but was finally launched here:

https://mcgill.ca/seedsofchange/project/mcgill-robotics

Thanks!