Learning from Demonstration Experiments

Code and recorded experiment data for the MSc thesis Learning from Demonstration for Robotic Arm Control: Trajectory Representations and Control Interfaces.

Contents

• lfd_v2/: corrected experiment implementation
• env/, experts/, models/: original environment, demonstrations, and model code
• tests/: unit and regression tests
• run_v2.py: main command-line entry point
• results_v2/comparison/: aggregate tables used in the thesis
• results_v2/compatible/ and results_v2/rigid_body/: per-episode CSV files, run manifests, and summaries

Model checkpoint files are excluded because of their size. Each retained manifest records the configuration, case identifiers, and configuration hash needed to identify a run.

Environment

Python 3.9 or later is recommended.

python -m pip install -r requirements.txt

Verification

python -m unittest discover -s tests

Running experiments

Run a small two-environment check:

python run_v2.py smoke

Run individual studies:

python run_v2.py main
python run_v2.py oracle
python run_v2.py limits
python run_v2.py freeze-plan
python run_v2.py ood
python run_v2.py smoothness
python run_v2.py scaling
python run_v2.py report

Experiments resume completed runs by default. Use --no-resume to rerun them.