Hi everyone,
We are nearing the end of the Final project.
Some last tips:
- The testing environment consisting of a 'familiar' part (i.e. the known training environments) and an unknown part (testing environment) involving a bridge and some object perturbations (see also the Takeshi's castle video below).
- The challenge is to move forward (in x-direction) on the bridge, centered around y=0
- You are free to adapt any part of the code base, including EA, robot description, observations, control method etc. As long as the experiment involves robot evolution.
- Start testing your solution(s) on time using:
python final_project_test.py --best_dir_path results/final_project- Make sure the testing script works
- Keep track of all the files necessary to run your code (e.g. different controller.py code or changed sensor classes etc.)
- When unsure, make a fresh clone of your fork and test if you can run the final_project_test script (remember to checkout and install the clean repo)
- Make use of the cluster to increase productivity: setup instructions here
- Parallelize evaluations to speed up evolution (remember parallelization of Challenge 3)
- Run long experiments overnight
- Test different configuration of your experiment
- You can buffer multiple experiments in a queue and the cluster will run them when compute is available
- Final grading is based on how you designed your evolutionary experiment.
- Understand the problem your facing, and propose a solution
- What are the challenges of your idea and what do you do to solve it?
- Do the result support your hypothesis, why (not)?
The timeline for the next weeks
- 21st May: Submission on moodle (21st of May, 23.59).
- Instructions are provided on the submission page, but feel free to contact any of the TA's if something goes wrong
- 28th May: Presentation (attendance is mandatory)
Good luck with the final project and see you next week!
Fuda
Watch the Takeshi's Castle video
https://www.youtube.com/watch?v=V8D5sdDLYME
