synapsys 0.2.4

Modern Python control systems framework with distributed multi-agent simulation

Synapsys

Modern Python control systems framework with distributed multi-agent simulation

Documentation · Quickstart · PyPI · Examples · Changelog

---

Overview

Synapsys is an open-source Python library for modelling, simulating, and deploying control systems. It provides a MATLAB-compatible API built on SciPy, a modern multi-agent simulation framework, and a pluggable transport layer (shared memory / ZeroMQ) that scales from a single laptop to distributed lab setups.

Any PyTorch, Keras or JAX model plugs directly into a ControllerAgent via a plain np.ndarray -> np.ndarray callback, making it straightforward to combine classical control theory with deep learning or reinforcement learning.

from synapsys.api import tf, ss, feedback, step, c2d

# SISO
G    = tf([1], [1, 2, 1])    # G(s) = 1 / (s^2 + 2s + 1)
T    = feedback(G)            # closed-loop: T = G / (1 + G)
t, y = step(T)                # step response
Gd   = c2d(G, dt=0.02)       # ZOH discretisation at 50 Hz

# MIMO
G_mimo = tf([[[ 1], [0]],
             [[ 0], [1]]],
            [[[1,1],[1]],
             [[1],[1,2]]])
T_mimo = feedback(G_mimo)     # returns StateSpace closed-loop

---

Demo — Quadcopter MIMO Neural-LQR

12-state linearised quadcopter controlled by a residual Neural-LQR (du = -K*e + MLP(e)). The MLP output layer is zeroed at initialisation, so the controller starts as provably stable LQR and the residual can be trained later via RL or imitation learning without destabilising the loop.

PyVista 3D window — drone mesh, trajectory trail and live HUD at 50 Hz

matplotlib telemetry — x-y position, altitude, Euler angles and control deviations

Full walkthrough: Quadcopter MIMO Neural-LQR example

---

Features

Feature	Description
MATLAB-Compatible API	tf(), ss(), c2d(), step(), bode(), feedback(), lsim() — same names, pure Python
LTI Core	TransferFunction, StateSpace, and TransferFunctionMatrix with operator overloading, poles, zeros, stability
MIMO Support	TransferFunctionMatrix for multi-input multi-output plants, MIMO feedback(), transmission zeros via Rosenbrock pencil
Control Algorithms	Discrete PID with anti-windup, LQR via algebraic Riccati equation (Q/R validated)
AI Integration	Any PyTorch, Keras or JAX model as a controller — plain callable interface, no wrappers
Multi-Agent Simulation	PlantAgent and ControllerAgent with lock-step and wall-clock sync
Distributed Transport	Zero-copy shared memory (single-host) and ZeroMQ PUB/SUB and REQ/REP (multi-process / multi-machine)
Hardware Abstraction	HardwareInterface contract enables seamless MIL to SIL to HIL transitions
Matrix Builders	StateEquations, mat(), col(), row() — define state-space models from named equations

---

Installation

Requirements: Python >= 3.10, NumPy >= 1.24, SciPy >= 1.10, pyzmq >= 25.0

# pip
pip install synapsys

# uv
uv add synapsys

# Poetry
poetry add synapsys

# conda / mamba  (conda-forge)
conda install -c conda-forge synapsys

For 3D visualisation (quadcopter example):

pip install synapsys[viz] torch matplotlib

For development:

git clone https://github.com/synapsys-lab/synapsys.git
cd synapsys
uv sync --extra dev

---

Quickstart

1. LTI systems and frequency analysis

from synapsys.api import tf, ss, bode, feedback, c2d

# SISO second-order transfer function
G = tf([1], [1, 2, 1])

# Closed-loop with unity negative feedback
T = feedback(G)

# Frequency response
w, mag, phase = bode(G)

# ZOH discretisation at 50 Hz
Gd = c2d(G, dt=0.02)

# MIMO: 2x2 transfer-function matrix
G_mimo = tf(
    [[[1], [0]], [[0], [1]]],
    [[[1, 1], [1]], [[1], [1, 2]]],
)
T_mimo = feedback(G_mimo)    # returns StateSpace

2. Control algorithms

from synapsys.algorithms import PID, lqr
import numpy as np

# Discrete PID with anti-windup saturation
pid = PID(Kp=3.0, Ki=0.5, Kd=0.1, dt=0.01, u_min=-10.0, u_max=10.0)
u   = pid.compute(setpoint=5.0, measurement=y)

# LQR — solves the algebraic Riccati equation
A = np.array([[0., 1.], [-2., -3.]])
B = np.array([[0.], [1.]])
K, P = lqr(A, B, Q=np.eye(2), R=np.eye(1))
# Control law: u = -K * x

3. State-space from named equations

from synapsys.utils import StateEquations

m, c, k = 1.0, 0.1, 2.0

eqs = (
    StateEquations(states=["x1", "x2", "v1", "v2"], inputs=["F"])
    .eq("x1", v1=1).eq("x2", v2=1)
    .eq("v1", x1=-2*k/m, x2=k/m,  v1=-c/m)
    .eq("v2", x1=k/m,  x2=-2*k/m, v2=-c/m, F=1/m)
)

print(eqs.A)   # 4x4 system matrix
print(eqs.B)   # 4x1 input matrix

4. Multi-agent closed-loop simulation

from synapsys.api import ss, c2d
from synapsys.agents import PlantAgent, ControllerAgent, SyncEngine, SyncMode
from synapsys.algorithms import PID
from synapsys.transport import SharedMemoryTransport
import numpy as np

# Discretise G(s) = 1/(s+1) at 100 Hz
plant_d = c2d(ss([[-1]], [[1]], [[1]], [[0]]), dt=0.01)

with SharedMemoryTransport("demo", {"y": 1, "u": 1}, create=True) as bus:
    bus.write("y", np.zeros(1))
    bus.write("u", np.zeros(1))

    pid  = PID(Kp=4.0, Ki=1.0, dt=0.01)
    def law(y: np.ndarray) -> np.ndarray:
        return np.array([pid.compute(setpoint=3.0, measurement=y[0])])

    sync = SyncEngine(SyncMode.WALL_CLOCK, dt=0.01)

    PlantAgent("plant", plant_d, bus, sync).start(blocking=False)
    ControllerAgent("ctrl",  law,  bus, sync).start(blocking=True)

5. MIL to SIL to HIL — swap transport, keep algorithm

# MIL — shared memory, single host
from synapsys.transport import SharedMemoryTransport
bus = SharedMemoryTransport("demo", {"y": 1, "u": 1}, create=True)

# SIL — ZeroMQ, cross-process or cross-machine
from synapsys.transport import ZMQTransport
bus = ZMQTransport("tcp://localhost:5555", mode="pub")

# HIL — real hardware
from synapsys.agents import HardwareAgent
from synapsys.hw import MockHardwareInterface
hw    = MockHardwareInterface(n_inputs=1, n_outputs=1, plant_fn=my_hw)
agent = HardwareAgent("hw", hw, bus, sync)

# PlantAgent and ControllerAgent stay exactly the same in all three modes

---

Examples

Example	Description
basic/step_response.py	Step response of a 2nd-order system
distributed/plant.py + controller.py	Two-process PID loop via shared memory
distributed/plant_zmq.py	Same loop over ZeroMQ (cross-machine)
advanced/01_custom_signals.py	Custom reference signals with lsim()
advanced/02_sil_ai_controller/02b_sil_ai_controller.py	SIL + Neural-LQR PyTorch controller on 2-DOF mass-spring-damper
advanced/03_realtime_scope.py	Text-mode real-time oscilloscope
advanced/04_realtime_matplotlib.py	Live matplotlib oscilloscope
advanced/05_digital_twin/05_digital_twin.py	Digital twin with mechanical wear detection
advanced/06_quadcopter_mimo/	12-state quadcopter MIMO Neural-LQR with PyVista 3D, config GUI, GIF export
quickstart_en.ipynb	Interactive Jupyter notebook walkthrough

---

Architecture

synapsys/
├── api/            # MATLAB-compatible facade  (tf, ss, c2d, step, bode, feedback, ...)
├── core/           # LTI math — TransferFunction, StateSpace, TransferFunctionMatrix, LTIModel
├── algorithms/     # PID (discrete, anti-windup), LQR (ARE solver)
├── agents/         # PlantAgent, ControllerAgent, HardwareAgent, SyncEngine
├── broker/         # MessageBroker, Topic, SharedMemoryBackend, ZMQBrokerBackend
├── transport/      # SharedMemoryTransport, ZMQTransport, ZMQReqRepTransport
├── hw/             # HardwareInterface (abstract) + MockHardwareInterface
└── utils/          # StateEquations, mat(), col(), row()

The transport layer is the key abstraction: agents communicate exclusively through a TransportStrategy interface. The broker/ module adds a higher-level pub/sub bus (backed by shared memory or ZMQ) for multi-agent scenarios. Swapping the concrete transport or broker backend requires changing one line — algorithms and agents are untouched.

---

Testing

uv run pytest                      # run all tests
uv run pytest --cov=synapsys       # with coverage report
uv run mypy synapsys               # type checking
uv run ruff check synapsys tests   # linting

Metric	Value
Test suite	287 tests
Coverage	100%
Type checking	mypy strict — 0 errors
Pre-commit hooks	ruff lint + format, mypy, pytest
Python versions	3.10, 3.11, 3.12

---

Roadmap

Version	Status	Features
v0.1.0	Released	SISO LTI, PID, LQR, multi-agent, shared memory, ZMQ, hardware abstraction, Neural-LQR example
v0.2.0	Released	MIMO support — TransferFunctionMatrix, MIMO feedback(), transmission zeros (Rosenbrock pencil), LQR Q/R validation, covariant type annotations
v0.2.1	Released	Quadcopter MIMO Neural-LQR example with PyVista 3D, config GUI, GIF export, version sync fix
v0.2.2	Released	MessageBroker pub/sub bus, 100% test coverage, mypy strict, pre-commit hooks
v0.3	Planned	margin(), rlocus(), pole_placement(), Kalman filter, Luenberger observer
v0.5	Planned	Real hardware drivers (serial, CAN, FPGA via PYNQ)

See CHANGELOG.md for the full release history.

---

Citing

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

BibTeX

@software{synapsys2026,
  author    = {Farias, Oseias D. and contributors},
  title     = {Synapsys: A Python Framework for Modelling and
               Real-Time Simulation of Linear Control Systems},
  year      = {2026},
  url       = {https://github.com/synapsys-lab/synapsys},
  license   = {MIT},
}

APA

Farias, O. D., & contributors. (2026). Synapsys: A Python framework for
modelling and real-time simulation of linear control systems.
https://github.com/synapsys-lab/synapsys

For the version number, use the version shown on PyPI.

---

Contributing

Contributions are welcome. Please open an issue to discuss what you would like to change before submitting a pull request.

git clone https://github.com/synapsys-lab/synapsys.git
cd synapsys
uv sync --extra dev
uv run pre-commit install   # install git hooks (ruff, mypy, pytest)
uv run pytest               # make sure everything passes before you start

---

Contributors

---

License

MIT © 2026 Synapsys Contributors