Bartłomiej Mielnicki

> @bmielnicki just one more question, is this supposed to work with the [trajectory visualization PR](https://github.com/HumanCompatibleAI/overcooked_ai/pull/38)? Currently, there is no integration between this PR and https://github.com/HumanCompatibleAI/overcooked_ai/pull/38 except they use the...

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...

Extension of this PR is https://github.com/HumanCompatibleAI/overcooked-demo/pull/30 It is separate PR because it requires review of https://github.com/HumanCompatibleAI/overcooked_ai/pull/38 and I don't want to prevent merge of previous changes because of that.

There is an upcoming [ipython notebook](https://github.com/HumanCompatibleAI/overcooked_ai/blob/ae9bd975f13b3e6d48c02468195ec29e293362f9/src/overcooked_ai_py/tutorial_notebooks/Agent_evaluator_intoduction.ipynb) on trajectory visualization branch that can be used as a good starting point to this task. What can/should be done with this notebook: 1....

Thanks for the suggestions! I will create a notebook with tutorial for current master branch code as my next task including those suggested things. I'm not familiar with human_aware_rl yet...

I've done a simple introduction notebook that works with the current master branch. https://colab.research.google.com/github/bmielnicki/overcooked_ai/blob/introduction_notebook/introduction.ipynb Some things worth to mention: - evaluate_human_model_pair method consistently produces 0 rewards, it is even worse...

Thanks for the reply! 1. Month ago or so (probably before the big changes in planners code) human model gave better results that 0 rewards. I'm not sure what happened...

You are right - agents are not pressing cook. I'm already close to pushing the fix for this bug,

Given that GreedyHumanModel accepts only situation when all_orders has len of 1 we can just cook any soup of selected len. inside ml_action I've changed this ``` if soup_nearly_ready and...

Features were not equal `self.assertTrue(np.array_equal(expected_featurization, featurized_observations))`. I've overwritten pickled features (by uncommenting line with save pickle), result is pushed to https://github.com/bmielnicki/overcooked_ai/tree/greedy_human_model_fix