DR Control

This is the repo for our paper Distributionally Robust Path Integral Control. Please kindly cite our work as following:

@INPROCEEDINGS{10644179,
  author={Park, Hyuk and Zhou, Duo and Hanasusanto, Grani A. and Tanaka, Takashi},
  booktitle={2024 American Control Conference (ACC)}, 
  title={Distributionally Robust Path Integral Control}, 
  year={2024},
  volume={},
  number={},
  pages={1164-1171},
  keywords={System dynamics;Heuristic algorithms;Diffusion processes;Optimal control;Data collection;Trajectory;Complex systems;stochastic optimal control;distributionally robust optimization;path integral method;risk-sensitive control},
  doi={10.23919/ACC60939.2024.10644179}}

Results and Visualizations

Performance of different schemes

Model	Scheme	Success Rate (%)	Arrive Time (s) Mean	Std. Dev.	95 prct.
Double Integrator	PIC	66	21.30	12.63	44.42
Double Integrator	DRPI	96	8.04	2.25	12.32
Unicycle	PIC	19	25.89	11.13	44.52
Unicycle	DRPI	78	16.94	6.17	28.30

How to Use

Requirement

conda env create -f environment.yml
conda activate DRC

To run the experiments

To run the script, use the following command:

python src/main.py --DR_method "DR NM" --Experiment "1" --Online --seed_value 1 --num_simulation 1 --observations 1

Options:

• --DR_method: Choose a DR method. Options are "DR NM", "DR GM", and "RN". Default is "DR NM".

  • "DR NM" represents "DRPI Numerical Minimization".
  • "RN" represents "Risk-Neutral PIC".

• --Experiment: Set the experiment number. Choose between "1" and "2". Default is "1".

  • "1" represents "Input Integrator".
  • "2" represents "Unicycle".

• --Visualization: If included, this will visualize the path integral algorithm for each simulation.

• --Online: If included. this will enable online parameter estimation.

• --seed_value: Set the seed value. Default is not setting seed.

• --num_simulation: Set the number of simulations. Default is 100.

• --observations: Set the number of observations. Default is 1.

• --sigma: Set the value of Sigma. Default is 0.5.

• --mu: Set the mu value. Default is [0.0, 0.0].

• --max_steps: Set the maximum number of steps. Default is 1000.

• --num_trajs: Set the number of trajectories. Default is 500.

• --num_vis: Set the number of vis. Default is 500.

• --T: Set the T value. Default is 2.0.

• --dt: Set the dt value. Default is 0.05.

To visualize the results from logs

To run the script, use the following command:

python src/vis.py --Experiment "1" --Name "log/DRExp1" --Visualization

Options:

• --Experiment: Choose a model number. Options are "1" and "2". Default is "1".

  • "1" represents "Input Integrator",
  • "2" represents "Unicycle".

• --Name: Set the file name. Options include "log/DRExp1", "log/DRExp2", "log/RNExp1", and "log/RNExp2". Default is "log/DRExp1".

• --SAVE_LOG: If included, this will save the figure.

• --Visualization: If included, this will visualize the whole simulation.