DT-MPC

Implementation of differentiable tube-based MPC controller based on the 2024 Robotics: Science and Systems paper Differentiable Robust Model Predictive Control.

Initial repository structure

• /dynamics: base plant dynamics and safety-embedded dynamics wrappers
• /solvers: optimal control solver interfaces
• /learning: DOC engine and DT-MPC training loop interfaces
• /scripts: runnable entry points
• /tests: smoke tests for the initial package structure

Dependencies

• numpy==2.4.4

Setup

Before running the script, setup the repo using:

cd DT-MPC pip install -e .

Need to add the conda environment as an exception to Windows Security > Virus & thread protection > Exceptions because of JAX.

test the basic dubin scene

python .\tests\test_dubin_trajectory.py

test the embedded state scene

python .\tests\test_embedded_state.py