Skip to main content

Distributed behavioral experiments

Project description

PyPI PyPI pyversions Documentation Status PyPI - Status

License: MPL 2.0 Contributor Covenant

Autopilot

Autopilot Banner Logo

Docs Paper Forum Hardware
Read the Docs Paper Forum Hardware (Coming Soon!)

Autopilot is a Python framework for performing complex, hardware-intensive behavioral experiments with swarms of networked Raspberry Pis. As a tool, it provides researchers with a toolkit of flexible modules to design experiments without rigid programming & API limitations. As a vision, it dreams of bridging the chaotic hacky creativity of scientific programmers with a standardized, communally developed library of reproducible experiment prototypes.

Autopilot was developed with three primary design principles:

  • Flexibility - Autopilot was designed for any hardware and any experiment -- its hardware API is designed to give a structured wrapper around the code you already use, and its task design is entirely non-prescriptive. It attempts to eliminate the need for researchers to use a patchwork of mutually incompatible tools to perform complex experiments. Autopilot is a hacker's plaything -- rather than a uniform, simplified experience, its modular design and complete API-level documentation is meant to encourage users to make and break core Autopilot modules.
  • Efficiency - Autopilot uses Python as a glue around high-performance, low-level libraries, and is fully concurrent across multiple threads, processes, and computers. Its distributed design eliminates the hard limits faced by by single-computer systems, letting researchers use arbitrary numbers and combinations of hardware components to perform complex, hardware-intensive experiments at scale.
  • Reproducibility - Autopilot obsessively documents data provenance, logging the entire history of an Animal's training, including any version and local code changes. Any part of an experiment that isn't documented is considered a bug. By integrating experiments and producing data that is clean at the time of acquisition, Autopilot makes it easy to do good science -- its goal is to allow exact experimental replication from a single file.

Distributed Behavior

Autopilot's premise is simple: to scale experiments, just use more computers.

Autopilot systems consist of multiple "Agents" -- computers with specialized roles in the swarm. One user-facing "Terminal" agent allows a researcher to control many "Pilots," or computers that perform experiments (typically the beloved Raspberry Pi). Each Pilot can coordinate one or many "Children" to offload subsets of an experiment's computational or hardware requirements. Users can use and misuse Autopilot's flexible modules to make whatever agent topology they need <3.

Autopilot System Diagram

Module Overview

Autopilot divides the logical structure of experiments into independent1 modules:

Module
Hardware Agents - Pilot & Terminal Runtime classes that encapsulate a computer/Pi's role in the swarm. Terminals provide the user interface and coordinate subjects and tasks, Pilots do the experiments. Formalizing the Agent API to allow additional agents like Compute or Surveillance agents is a major short-term development goal!
Hardware Hardware - Control your tools! Extensible classes to control whatever hardware you've got.
Hardware Stimuli - Stimulus management and presentation. Parametric sound generation with a realtime audio server built on Jackd. Stubs are present for future development of visual stimuli using Psychopy.
Hardware Tasks - Build experiments! Write some basic metadata to describe data, plots, and hardware and the rest is up to you :)
Hardware Subject - Data management with hdf5 and pyTables. Abstraction layer for keeping obsessive records of subject history and system configuration
Hardware Transforms - Composable data transformations. Need to control the pitch of a sound with a video? build a transformation pipeline to connect your objects
Hardware UI - UI for controlling swarms of Pilots using Qt5/PySide2
Hardware Visualization - (Mostly Prototypes) to do common visualizations

1 a continual work in progress!

Getting Started

All documentation is hosted at https://docs.auto-pi-lot.com

Installation is simple, just install with pip and use Autopilot's guided setup to configure your environment and preferences. The initial setup routine uses a CLI interface that is SSH friendly :)

pip3 install auto-pi-lot
python3 -m autopilot.setup.setup_autopilot

Autopilot Setup Console

All of Autopilot is quite new, so bugs, incomplete documentation, missing features are very much expected! Don't be shy about raising issues or asking questions in the forum.

What's new?

v0.3.0

After much ado, we're releasing Autopilot's first major upgrade. Cameras, Continuous data, DeepLabCut, and a lot more!

  • Autopilot has moved to Python 3!! (Tested on 3.6-3.8)
  • Capturing video with OpenCV and the Spinnaker SDK is now supported (See autopilot.hardware.cameras)
  • An I2C_9DOF motion sensor and the MLX90640 temperature sensor are now supported.
  • Timestamps from GPIO events are now microsecond-precise thanks to some modifications to the pigpio library
  • GPIO output timing is also microsecond-precise thanks to the use of pigpio scripts, so you can deliver exactly the reward volumes you intend <3
  • Hardware modules have been refactored into their own module, and have been almost wholly rebuilt to have sensible inheritance structure.
  • Networking modules are more efficient and automatically compress arrays (like video frames!) on transmission. Streaming is also easier now, check out Net_Node.get_stream() !
  • We now have a detailed development roadmap , so you can see the magnificent future we have planned.
  • We have created the autopilot-users discussion board for troubleshooting & coordinating community development :)

What's next?

Autopilot Development Todo

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for auto-pi-lot, version 0.3.1
Filename, size File type Python version Upload date Hashes
Filename, size auto-pi-lot-0.3.1.tar.gz (21.1 MB) File type Source Python version None Upload date Hashes View

Supported by

Pingdom Pingdom Monitoring Google Google Object Storage and Download Analytics Sentry Sentry Error logging AWS AWS Cloud computing DataDog DataDog Monitoring Fastly Fastly CDN DigiCert DigiCert EV certificate StatusPage StatusPage Status page