Relign is a Python software for simulating the active alignment of optical systems. More precisely, Relign can be used to analyze the sensor output of multi-lens systems by changing the relative alignment to the sensor. It also offers an interface for reinforcement learning algorithms to well-known open source libraries such as Gymnasium and Stable-Baselines.

An example of a trained RL agent, aligning a lens system with two lenses within 10 steps can be seen here:

The methodology and results are detailed in our paper, that is currently under review and available as a preprint on arXiv:2503.02075.

Ensure you have the following installed:

• Python 3.10+
• pip

Once cloned, install the package with:

pip install .
pip install .[dev]

pytest

Run

mkdocs serve

to build and serve the docs locally.

To familiarize yourself with using the environment for your own algorithms, see docs/getting_started.ipynb.

To start training a model, use the scripts/train.py script. You can customize the training parameters using command-line options.

python3 scripts/train.py --env="la" --model=PPO \
    --learning-rate=1e-3 --ent-coef=0.01 \
    --benchmark="b_OD_N000_256"

@misc{burkhardt2025relign,
      title={Active Alignments of Lens Systems with Reinforcement Learning}, 
      author={Matthias Burkhardt and Tobias Schmähling and Pascal Stegmann and Michael Layh and Tobias Windisch},
      year={2025},
      eprint={2503.02075},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2503.02075}, 
}

This work was funded by the German Federal Ministry of Research, Technology and Space (BMFTR) under grant number 13FH605KX2.