biosym

Installation:

For a standard install from PyPI:

1. Install IPOPT and ensure cyipopt can build or link against it: https://cyipopt.readthedocs.io/en/stable/install.html
2. Install biosym:

pip install biosym

If you want the iteration dashboard features, install the dashboard extra:

pip install "biosym[dashboard]"

For local development in this repository, use:

uv sync --group dev

Usage

biosym is a Python toolbox for building biomechanical models, exploring their structure and dynamics, and solving movement optimization problems. The examples cover three core workflows: loading and inspecting models for forward computations, setting up predictive gait optimal control problems from configuration files, and batching model evaluations with JAX for high-throughput or learning-based applications.

Load and inspect a model:

from biosym.model.model import load_model

# Load a simple pendulum model and rebuild the cached functions from source.
model = load_model("tests/models/pendulum.xml", force_rebuild=True)

# Inspect the generalized coordinates, speeds, and overall problem size.
print(model.coordinates["names"])
print(model.speeds["names"])
print(model.n_states, model.n_constants)

Set up and solve an optimal control problem from a YAML file:

from biosym.ocp import collocation

# Read model, objectives, constraints, and solver settings from YAML.
ocp = collocation.Collocation("examples/standing2d.yaml")
# Solve the optimal control problem and open the default visualization.
solution = ocp.solve(visualize=True)

Batch model evaluations with JAX:

import jax

from biosym.model.model import load_model
from biosym.utils import states

# Load a gait model with contact and actuator dynamics.
model = load_model("tests/models/gait2d_torque/gait2d_torque.yaml")
# Replicate the default input state into a batch for vectorized evaluation.
batched_inputs = states.stack_dataclasses([model.default_inputs] * 128)

# Vectorize the model constraint function across the batch dimension.
dynamics_fn = jax.vmap(model.run["confun"], in_axes=(0, None))
batched_output = dynamics_fn(batched_inputs.states, batched_inputs.constants)

# The result contains one dynamics evaluation per batch element.
print(batched_output.shape)

Near-term roadmap

We are actively developing biosym and welcome contributions from the community. Currently, we work on making it easier to install and use, which also means that some structural changes should be expected. Contact us if you want to contribute or have suggestions for improvements.

Additionally, here is what to expect from biosym in the near future:

1. Deep learning integration.
2. Differentiability with respect to musculoskeletal model parameters.
3. 3D-ready optimal control simulations.
4. Continued documentation improvements.

Follow our group's LinkedIn page for updates on new features and releases.

Citation

If you use biosym in academic work, please cite it as software:

@software{biosym_2026,
  title = {biosym: A Python toolbox for biomechanical movement simulation and optimal control},
  author = {Markus Gambietz and Theodoros Balougias and Asghar Mahmoudi and Yipeng Zhang and Anne Koelewijn},
  year = {2026},
  version = {0.1.0},
  url = {https://github.com/biomac-group/biosym}
}