Deflated Dynamics Value Iteration

This is the official implementation for the paper Deflated Dynamics Value Iteration at TMLR 2025.

DDVI Experiments

To run the experiment for Figure 1, set the environment config in the files planning/exp_DDVI.py and planning/plotter_ddvi. Then run:

python planning/exp_DDVI.py
python planning/plotter_ddvi.py

To run experiments for Figure 2, run

python planning/exp_error_comparison.py
python planning/exp_horizon_comparison.py
python planning/exp_size_comparison.py
python planning/plotter_comparison.py

DDTD Experiments

For Figure 3, run the following in the sample_based directory:

python exp_pe_sample.py config_maze55_0.95.yaml ALL --num_trials 20
python exp_pe_sample.py config_chainwalk_0.95.yaml ALL --num_trials 20

For plotting, use plotter_pe_main.py and plotter_pe_multimodel_error.py. The path to data should be set in the exp_dir variable.