named parameters in up.sh; saving trajectories; trajectories replay

[bmielnicki]

1. Added usage of named parameters to up.sh. They can be used everything that was specified by env variables (that are no longer used), also trajectories and agents directories.
2. Pass to docker-compose all env variables by .env file. This prevents any naming collisions with other applications.
3. Trajectory and agent directories are now passed as volumes so all changes there (like saving trajectory) are now kept between sessions.
4. Added saving trajectories to trajectory directory. It uses OvercookedEnv so there is no code duplication between overcooked-demo and overcooked_ai code. Checkbox in index.html is added so user can choose if trajectory should be saved.
   
5. Small changes in OvercookedPsiturk to work with changes in OvercookedGame.

I plan to add additional stuff into this branch (like replaying trajectories from trajectory directory), but changes above seems to be mergable into master now.

[mesutyang97]

Wow! Thanks for adding this long-awaited feature! LGTM but would like to get feedback from @nathan-miller23

I have cloned this branch and tested it locally. It does indeed work

[mesutyang97]

@bmielnicki just one more question, is this supposed to work with the trajectory visualization PR?

[bmielnicki]

@bmielnicki just one more question, is this supposed to work with the trajectory visualization PR?

Currently, there is no integration between this PR and HumanCompatibleAI/overcooked_ai#38 except they use the same trajectory format. I plan to get them both working together - when the player picks the trajectory file to visualize he will be able to see both the replay and the chart.

[bmielnicki]

Additional changes are pushed:

1. Creating trajectory with selected filename (with use of template values for layout name or timestamp)
2. Replay of trajectories
   • based OvercookedTrajectoryReplay on the legacy branch
   • currently, it is not an object as I've had problems with changing context when putting all variables and functions into class (this changes context from object to window easily)
   • Timesteps per render number is used for controlling how many timesteps should be between consecutive renders (with high Timesteps per second variable whole code is slowing down if we want to render every state).
     

Next step:

• Use the code from Trajectory visualization overcooked_ai#38 to render chart below the game (and line up slider handle with timesteps on the chart). Add a checkbox to turn on and off the chart and/or trajectory replay.

As before - this change is ready to be merged after getting feedback.

[bmielnicki]

Extension of this PR is #30 It is separate PR because it requires review of HumanCompatibleAI/overcooked_ai#38 and I don't want to prevent merge of previous changes because of that.