Actively Learning Skills to Prepare a Human's Preferred Meal

Maintainer: Pat Callaghan
Last updated: March 11th, 2022

Expected outputs

Upon executing the main program, one should see terminal outputs indicating the human's desired parameters and weights. Next, one should see indications of the following Python objects' successful instantiation:

• An ideal pizza
• A particle filter
• A set of particles
• A query generator

Subsequent terminal outputs should display the expected parameter values and importance weights of the learned reward function after each (simulated) query. One will also see terminal outputs indicating when the particle filter resamples its set of particles.

At the end of a successful program execution, one should see a plot of a hypothesized pizza generated according to the learned reward model when it was most similar to the true model at any point during execution. Depending on the feasibility of the learned model's parameters and the error threshold one specifies for generating a pizza, generation can take several minutes.

Running the code

Installation

From the top level of the learning_skills_from_humans directory, execute the following from the commandline:

conda env create -f learning_skills_from_humans.yml

One can optionally modify the primary module to be executable:

sudo chmod +x robot_chef_program.py

Main program

From within the top level of the learning_skills_from_humans directory, execute the following from the commandline:

python robot_chef_program --<flag_1> <arg_1> ... --<flag_n> <arg_n>

And if one opted to make the primary module executable:

./robot_chef_program --<flag_1> <arg_1> ... --<flag_n> <arg_n>

Please find the arguments and their corresponding flags in the arguments.py file located within the utils/ sub-directory.

Data Visualization

Upon successful program execution, one can locate the corresponding data in the pertinent csv files found within the data/ sub-directory.

To visualize the results, cd into the utils/ sub-directory and execute the following from the commandline:

./create_data_plots.py ../data/<pertinent sub-directory>/<pertinent file>.csv \
 --context <desired plot context>

with current eligible contexts:

• expected_values
• choice_comparisons