newton-actuators

What it is and what it does

Newton Actuators is a GPU-accelerated library for computing control forces and torques in physics simulations. It provides seven actuator classes ranging from simple stateless PD controllers to stateful PID, delayed PD, DC motor models, and neural network actuators (MLP and LSTM). Each actuator reads from simulation state arrays and writes computed forces back to control arrays, with support for both CUDA-graphable and non-graphable execution paths.

The library is built on top of warp-lang for GPU acceleration and optionally integrates PyTorch for neural network actuators. However, the package is now deprecated: starting with Newton 1.3, all actuators have been moved into the main newton package as newton.actuators, and this standalone distribution is no longer maintained. Existing users should migrate to the built-in module.

Use it for:

• Implementing PD or PID feedback control for robotic joint actuators in GPU-accelerated physics simulations.
• Modeling DC motor dynamics with velocity-dependent torque saturation in physics pipelines.
• Using learned neural network policies (MLP or LSTM) as actuators to drive simulation control.
• Parsing actuator definitions from USD files and instantiating the appropriate control law.
• Running stateful control loops with double-buffered state management in CUDA graphs.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Provides GPU-accelerated actuator implementations (PD, PID, DC motor, neural network) that compute forces and torques for physics simulations by reading simulation state and writing control outputs.

No—do not install. The package is explicitly deprecated and unmaintained as of Newton 1.3. The same functionality is now available in the main newton package as newton.actuators. Migrate to that instead to receive ongoing maintenance and updates.

Install

newton-actuators on PyPI

pip install newton-actuators

uv add newton-actuators

poetry add newton-actuators

Installing newton-actuators

Before you install

Low friction wheel install. Package is marked inactive (Development Status 7) and explicitly deprecated in favor of the built-in newton.actuators module starting with Newton 1.3; no longer maintained.

License in practice

Apache-2.0 permissive license allows commercial and private use with attribution; no restrictions on derivative works or distribution.

Quickstart

pip install newton-actuators

import warp as wp
from newton_actuators import ActuatorPD

indices = wp.array([0, 1, 2], dtype=wp.uint32)
pd = ActuatorPD(
    input_indices=indices,
    output_indices=indices,
    kp=wp.array([100.0, 100.0, 100.0], dtype=wp.float32),
    kd=wp.array([10.0, 10.0, 10.0], dtype=wp.float32),
    max_force=wp.array([50.0, 50.0, 50.0], dtype=wp.float32),
)
pd.step(sim_state, sim_control, None, None, dt=0.01)

Requires warp-lang as a runtime dependency; CUDA 12.x or 13.x for GPU execution; PyTorch optional but required for ActuatorNetMLP and ActuatorNetLSTM.

Verify before relying

• Whether existing code using newton-actuators will continue to work with future Newton releases or if migration is urgent.
• Compatibility details between this 0.1.1 release and the newton.actuators module it is being superseded by.
• Whether PyTorch extras (torch-cu12, torch-cu13) are still available or have been removed.

Package facts

Evidence: newton_actuators-0.1.1-py3-none-any.whl

Tags

More Scientific/Engineering packages

Further reading