scripting tools for streamlined laboratory automation

pip install labbench==0.20.1



This is a set of python tools for writing laboratory automation scripts that are clear, concise, and explainable. Code that achieves these goals should read like a pseudocode expression of the experimental procedure. Objects for control over equipment (or other software) should only expose a clear set of automation capabilities to make laboratory automation more robust and less frustrating.

The labbench module provides tools that support toward this goal through an object protocol and support functions. These separate repetitive and error-prone boilerplate code, Use of these capabilities among multiple experimental runs also helps to produced data sets with consistent structure.


A Device object exposes automation control over a piece of lab equipment, or software as a virtual "device." Organizing access into the Device class immediately provides transparent capability to

  • log data from the device
  • define consistent coercion between pythonic and over-the-wire data types
  • apply value constraints on instrument parameters
  • support threaded operation concurrent I/O
  • hook the Device state to user interface display
  • ensure device disconnection on python exceptions

Typical Device driver development work flow focuses communicating with the instrument. The drivers are made up of descriptors and methods, thanks to a small, targeted set of convenience tools focused on data types and communication backends. The following VISADevice backend illustrates a complete example on a complete power sensor:

import labbench as lb
import pandas as pd

class PowerSensor(lb.VISADevice):
    SOURCES = 'IMM', 'INT', 'EXT', 'BUS', 'INT1'
    RATES = 'NORM', 'DOUB', 'FAST'

    initiate_continuous ='INIT:CONT')
    output_trigger ='OUTP:TRIG')
    trigger_source ='TRIG:SOUR', only=SOURCES, case=False)
    trigger_count ='TRIG:COUN', min=1, max=200)
    measurement_rate ='SENS:MRAT', only=RATES, case=False)
    sweep_aperture ='SWE:APER', min=20e-6, max=200e-3, help='time (s)')
    frequency ='SENS:FREQ', min=10e6, max=18e9, step=1e-3,
                         help='input signal center frequency (in Hz)')

    def preset(self):

    def fetch(self):
        """ return a single power reading (if self.trigger_count == 1) or pandas Series containing the power trace """
        response = self.query('FETC?').split(',')
        if len(response) == 1:
            return float(response[0])
            return pd.to_numeric(pd.Series(response))

The VISADevice backend here builds interactive traits (python descriptors) from the SCPI commands given in each key. This is a functioning instrument automation driver that works on an actual commercial instrument:

with PowerSensor('USB0::0x2A8D::0x1E01::SG56360004::INSTR') as sensor:
    # configure from scratch

    # the following set parameters on the power sensor
    sensor.frequency = 1e9
    sensor.measurement_rate = 'FAST'
    sensor.trigger_count = 200
    sensor.sweep_aperture = 20e-6
    sensor.trigger_source = 'IMM'
    sensor.initiate_continuous = True

    power = sensor.fetch()

The usage here is simple because the methods and traits for automation can be discovered easily through tab completion in most IDEs. They can be used on connection with a simple with block.

Scaling to testbeds

Large test setups can neatly organize procedures that require a few Device instances into Task objects. A Testbed class collects the set of Task instances needed to perform the experiment, manages connection of these devices together, ensuring graceful disconnection of all Device on unhandled exceptions. A multitask definition defines a more complete experiment in the Testbed as a concurrent and sequential steps from multiple Task objects, using multitask. The Testbed also optionally exposes device state and fetched data for database management and user interface. The following ties all these together:

import labbench as lb
from myinstruments import MySpectrumAnalyzer, MySignalGenerator # custom library of Device drivers

class Synthesize(lb.Rack):
    inst: MySignalGenerator

    def setup(self, *, center_frequency):
        self.inst.center_frequency = center_frequency
        self.inst.bandwidth = 2e6

    def arm(self):
        self.inst.rf_output_enable = True

    def finish(self):
        self.inst.rf_output_enable = False

class Analyze(lb.Rack):
    inst: MySpectrumAnalyzer

    def setup(self, *, center_frequency):
        self.inst.center_frequency = center_frequency

    def acquire(self, *, duration):

    def fetch(self):
        # testbed data will have a column called 'spectrogram', which
        # point to subdirectory containing a file called 'spectrogram.csv'
        return dict(spectrogram=self.inst.fetch_spectrogram())

db = lb.SQLiteLogger(
    'data',                         # path to a new directory to contain data
    dirname_fmt='{id} {host_time}', # format string for relational data
    nonscalar_file_type='csv',      # numerical data format
    tar=False                       # True to embed relational data in `data.tar`

sa = MySpectrumAnalyzer(resource='a')
sg = MySignalGenerator(resource='b')

# tasks use the devices
generator = Synthesize(inst=sg)
detector = Analyze(inst=sa)

procedure = lb.Sequence(
    setup=(generator.setup, detector.setup),  # concurrently execute the long setups
    arm=generator.arm, # arm the generator before the detector acquires
    acquire=detector.acquire, # start acquisition in the generator
    fetch=(generator.finish, detector.fetch), # concurrently cleanup and collect the acquired data
    finish=(db.new_row, db.write),          # start the next row in the database
) here exposes the general capabilities of an experimental setup. An experiment can be defined and run in a python script by sweeping inputs to procedure:


Testbed = lb.Rack._from_module('testbed')

with Testbed() as test:
    # flow inside this `with` block only continues when no exception is raised.
    # On Exception, devices and the database disconnect cleanly.

    for freq in (915e6, 2.4e9, 5.3e9):
        ) # each {task}_{argname} applies to all methods in {task} that accept {argname}

This script focuses on expressing high-level experimental parameters, leaving us with a clear pseudocode representation of the experimental procedure. The test results are saved in an SQLite database, 'data/master.db'. Each row in the database points to spectrogram data in subdirectories that are formatted as 'data/{id} {host_time}/spectrogram.csv'.

Sometimes it is inconvenient to define the input conditions through code. For these cases, labbench includes support for tabular sweeps. An example, freq_sweep.csv, could look like this:

Step detector_center_frequency generator_center_frequency detector_duration
Condition 1 915e6 915e6 5
Condition 2 2.4e9 2.4e9 5
Condition 3 5.3e9 5.3e9 5

A command line call takes this table input and runs the same experiment as

labbench procedure freq_sweep.csv

This creates a testbed object from, and steps through the parameter values on each row of freq_sweep.csv.


Start in an installation of your favorite python>=3.7 distribution.

  • To install the current version, open a command prompt and type pip install labbench
  • To install the "stable" version (master branch) from git, open a command prompt and type pip install git+
  • To install the "bleeding edge" development version (develop branch) from git, open a command prompt and type pip install git+

If you plan to use VISA devices, install an NI VISA [1] runtime, such as this one for windows.


Getting started

Using drivers and labbench goodies for laboratory automation

Writing your own device driver

  • Introduction
  • VISA instruments
  • Serial port devices
  • .NET [1] library
  • Command line wrapper
  • Python module wrapper interface

Reference manuals


The following types of backend classes are implemented to streamline development of new instrumentation drivers:

  • ShellBackend (standard input/output wrapper for command line programs)
  • DotNet (pythonnet backend for dotnet libraries)
  • LabViewSocketInterface (for controlling LabView VIs via a simple networking socket)
  • SerialDevice (pyserial backend)
  • SerialLoggingDevice (pyserial backend for simple data streaming)
  • TelnetDevice (telnetlib backend)
  • VISADevice (pyvisa backend)
  • EmulatedVISADevice (test-only driver for testing labbench features)


Name Contact
Dan Kuester (maintainer)
Shane Allman
Paul Blanchard
Yao Ma

[1] Certain commercial equipment, instruments, or materials are identified in this repository in order to specify the application adequately. Such identification is not intended to imply recommendation or endorsement by the National Institute of Standards and Technology, nor is it intended to imply that the materials or equipment identified are necessarily the best available for the purpose.