dynibo 0.4.0

Tree-structured robot kinematics and dynamics with runtime-size workspace APIs
Documentation

dynibo is a fast, lightweight, and reliable library for robot kinematics and dynamics. It loads robot topology from URDF at runtime and provides allocation-free calculations through reusable per-robot storage. Python and C/C++ interfaces are available on top of the same Rust core.

Features

Fast

Across the benchmarks below, Dynibo runs 1.19–2.51× as fast as Pinocchio for the measured core operations. It is written in Rust and keeps allocation outside the calculation loop. After a Robot and output buffers are created, the main kinematics and dynamics routines reuse internal memory without allocating or resizing.

The table below shows Dynibo's speedup over Pinocchio for core kinematics and dynamics operations.

Model FK Jacobian Gravity RNEA
Two-leaf tree (7 joints, fixed base) 1.90× 2.05× 1.89× 1.94×
Two-leaf tree (7 joints, floating base) 2.16× 2.51× 2.15× 2.20×
Serial chain (40 joints, fixed base) 1.19× 1.49× 1.78× 1.99×
Serial chain (40 joints, floating base) 1.21× 1.56× 1.79× 2.09×

These Criterion quick-mode results use the same URDF models and joint states on an Intel Core i9-14900K with rustc 1.97.1 and Pinocchio 3.9.0. Setup and allocation are excluded, and speedups use interval medians after subtracting the measured 0.703 ns fixed C ABI overhead. With Pinocchio available through pkg-config, rerun the raw benchmarks with:

cargo bench --features pinocchio-bench --bench pinocchio -- --quick

Lightweight

Dynibo intentionally focuses on the most commonly used robot kinematics and dynamics interfaces:

  • forward_kinematics — target-link pose
  • jacobian — target-link Jacobian
  • jacobian_derivative — time derivative of the target-link Jacobian
  • forward_velocity_kinematics — spatial velocity
  • forward_acceleration_kinematics — spatial acceleration
  • inverse_kinematics — damped least-squares IK
  • mass_matrix — joint-space mass matrix
  • velocity_product_forces — Coriolis and centrifugal generalized forces
  • gravity — gravity compensation with optional external loads
  • inverse_dynamics — recursive Newton–Euler inverse dynamics
  • forward_dynamics — linear-time articulated-body forward dynamics

The API is built around a small set of types: Robot, LinkId, Frame, Twist, and Wrench. Rust, Python, C, and C++ interfaces share the same Rust implementation.

Reliable

Dynibo combines maintained fixtures with a seed-reproducible generated-URDF corpus, covering serial and branched robots, fixed and floating bases, mixed joint types, external loads, invalid inputs, and repeated workspace use. Results are checked against finite-difference approximations, consistency relations between related algorithms, and outputs from an independent Pinocchio oracle. Separate tests verify allocation-free execution and the installed Rust, Python, C, and C++ packages. See the test architecture for details.

Dependencies

The Rust core has two direct runtime dependencies:

Python wheels bundle the native library and have no runtime Python dependencies.

Quick start

Rust

Add the Cargo package:

cargo add dynibo

Load a URDF and compute a target-link pose:

use dynibo::Robot;

fn main() -> dynibo::Result<()> {
    let mut robot = Robot::from_urdf("robot.urdf")?;
    let tool = robot.link_id("tool")?;
    let q = vec![0.0; robot.joint_count()];

    let pose = robot.forward_kinematics(&q, tool)?;
    println!("translation: {}", pose.translation.vector.transpose());
    Ok(())
}

Python

Install the Python package from PyPI:

python -m pip install dynibo

The Python binding owns its reusable native calculation storage:

from dynibo import Robot

robot = Robot.from_urdf("robot.urdf")
tool = robot.link_id("tool")
q = [0.0] * robot.joint_count
pose = robot.forward_kinematics(q, tool)
print(pose.translation)

C/C++

C and C++ users can download a prebuilt package for Linux, macOS, or Windows from GitHub Releases, or build and install the package from source. Prebuilt packages contain the shared library, C and C++ headers, pkg-config metadata, and a CMake package configuration. Select the archive matching your operating system and CPU architecture and verify it against the release's SHA256SUMS.

Building from source requires Rust with Cargo and CMake 3.16 or newer:

cmake -S . -B build/c -DCMAKE_BUILD_TYPE=Release
cmake --build build/c --parallel
cmake --install build/c --prefix /opt/dynibo

Use an extracted prebuilt package or a source installation from another CMake project:

find_package(dynibo CONFIG REQUIRED)
target_link_libraries(my_robot PRIVATE dynibo::dynibo)

Configure the consumer with -DCMAKE_PREFIX_PATH pointing to the extracted archive directory or the installation prefix. See the installation guide for platform-specific runtime library paths.

Examples

Complete Rust, Python, C++, and C examples are available in the examples/ directory. Each example exercises all of the main kinematics and dynamics methods listed above.

Supported models

Dynibo supports runtime-sized tree URDFs with revolute, continuous, prismatic, and fixed joints. It rejects invalid topology and reports structured errors for bad input lengths, model-mismatched handles, and solver failures.

Testing

cargo fmt --all -- --check
cargo clippy --workspace --all-targets --locked -- -D warnings
cargo test --workspace --all-targets --locked

Run the complete local Rust, Python, C, and C++ verification suite with the command below. Pinocchio reference tests are included when Pinocchio is available through pkg-config.

bash ci/test-all.sh

Contributing

Dynibo is still at an early stage, and contributions are welcome. See CONTRIBUTING.md for development setup, required checks, and pull request guidelines.

Citation

If Dynibo is useful in your work, please cite it as:

@software{xue2026dynibo,
  author  = {Xue, Xiaojie},
  title   = {Dynibo: a Fast, Lightweight, and Reliable Robot Kinematics and Dynamics Library},
  year    = {2026},
  version = {0.4.0},
  url     = {https://github.com/xiaojie-xue/dynibo}
}