ACTOROB

Actuators Co-design and Task-aware Optimization for Robots

ACTOROB is a research codebase for trajectory optimization and actuator co-design in robot examples used to illustrate materials of a scientific article. The repository currently focuses on:

• trajectory optimization for configured task sets;
• actuator parameter modeling and evaluation;
• inverse-design loops built on top of Optuna + CMA-ES;
• report and dashboard generation for experiment outputs.

Project page: Task-Aware Actuator Parameter Allocation for Multibody Robots Paper DOI: 10.1109/LRA.2026.3674006

Release Roadmap

ACTOROB is currently in a pre-release stage. The current repository snapshot already covers the main pipeline on quadruped examples, while the broader release roadmap is:

• Main research pipeline available today on the quadruped examples included in this repository.
• Full public release with humanoid robot examples and documentation.
• Unified cross-robot workflow with an additional examples.

Current Support Status

The canonical local setup is pixi on osx-arm64 (Apple Silicon macOS). The release CI matrix also targets linux-64 so the same documented commands are exercised on both platforms before release.

The support policy and escalation path for unsupported environments are documented in SUPPORT.md.

Dependency Profiles

• Full trajectory-optimization and inverse-design workflows: use pixi. This is the supported path for the solver stack and research examples.
• Lightweight/reporting workflows via pip: the package exposes optional extras for reporting and terminal progress helpers, for example:

pip install -e ".[reporting,progress]"

This partial pip flow is useful for lighter API access and report generation, but it is not the primary supported installation path for the full research stack.

Quickstart

1. Install pixi.
2. Clone this repository.
3. Install the project environments:

pixi install --all

4. Run the test suite:

pixi run pytest -q

5. Run the quick inverse-design smoke test:

pixi run smoke-invdes

This smoke task runs a single-iteration walk example and writes its record into the system temp directory so the repository stays clean.

For a fuller inverse-design run, start from:

pixi run python examples/cma_es_invdes.py --config configs/dog_aligator_minimal.toml --tasks walk upstairs jump_forward

That longer run writes artifacts into outputs/ by default and can also emit an HTML dashboard.

Main Entry Points

actorob.invdes contains the inverse-design workflow used by the repository examples:

• ParallelAskTellOptimizer orchestrates batched ask/tell optimization.
• OptunaCmaEsStudyFactory provides the Optuna + CMA-ES backend.
• build_trajectory_bundle(...) wires inverse design to the trajectory optimizer and actuator model.

Useful commands during development:

pixi run pytest -q
pixi run python examples/cma_es_invdes.py --help
pixi run -e dev pre-commit-all
pixi run -e dev python -m build --sdist --wheel
pixi run -e dev check-dist
pixi run -e dev verify-wheel-install
pixi run -e dev verify-release
pixi run -e dev install-git-hooks

To enable the git hook locally so pre-commit runs automatically before each commit:

pixi run -e dev install-git-hooks

Repository Layout

• src/actorob/: library code.
• configs/: example experiment configurations.
• examples/: runnable entry points for demonstrations and smoke tests.
• robots/: robot models and meshes used by the examples.
• tests/: regression and API tests.

Open Source Notes

• The repository is being prepared for public release as a research artifact.
• CITATION.cff provides a software citation entry for the repository.
• CHANGELOG.md tracks release-facing changes starting from the public open-source preparation.
• SECURITY.md describes the current vulnerability disclosure path.
• SUPPORT.md documents supported environments and what maintainers expect in support requests.
• docs/REPRODUCIBILITY.md documents the currently validated commands and expected outputs.
• docs/ASSET_PROVENANCE.md documents the bundled public asset families and the generated files that should stay out of version control.

Acknowledgements

ACTOROB builds on a strong open robotics software stack. In particular, we would like to acknowledge:

• Aligator for constrained trajectory optimization components used by the motion-planning stack.
• Pinocchio for multibody kinematics and dynamics primitives used throughout the model and optimization pipeline.
• Meshcat for interactive visualization used in dashboards and report outputs.

Citation

If you use the scientific results behind this repository, please cite the article below. The repository-level software citation is also provided in CITATION.cff.

@ARTICLE{11433790,
  author={Nasonov, Kirill and Kakanov, Mikhail and Skvortsova, Valeria and Zaliaev, Eduard and Borisov, Ivan},
  journal={IEEE Robotics and Automation Letters}, 
  title={Task-Aware Actuator Parameter Allocation for Multibody Robots}, 
  year={2026},
  volume={11},
  number={5},
  pages={5869-5874},
  keywords={Actuators;Robots;Motors;Torque;Legged locomotion;Optimization;Friction;Costs;Topology;Humanoid robots;Actuators;humanoid robots;legged locomotion;motion planning},
  doi={10.1109/LRA.2026.3674006}}