Skip to content

Latest commit

 

History

History
112 lines (84 loc) · 5.22 KB

readme.md

File metadata and controls

112 lines (84 loc) · 5.22 KB
Raw

Jarvis

An AI powered home automation assistant in Rust

Jarvis is an Iron Man inspired pet project that provides a basic home automation assistant with customizable commands. It runs completely locally and doesn't need an internet connection. It knows how to turn various things (lights, appliances, ...) on or of, it can raise or lower window blinds, temperature and all sorts of other things. It is extensible and can accomodate any sort of command. It includes a quirky question answering capability powered by GPT2 that is more hilarious than useful.

I developed Jarvis solely out of interest in AI and Rust and learning more about both. It is a proof of concept automation assistant that isn't reliable but is very extensible and could easily be made improved.

Jarvis at work

Learning about Jarvis

I made an entire blog post series on how I implemented it and what I learned by doing it. It's meant to be a teaching project for other people that might be interested in a similar project. It can be found on my janhalozan.com website. The series starts with the intro post at: https://janhalozan.com/2024/07/07/home-assistant-intro/.

Installation

Jarvis runs completely locally but does require a few things to be installed on the system in order to run. I'd recommend checking out the blog post for details on how to install particular things.

To run Jarvis it needs:

  • A compatible pytorch version for the tch crate. I'm using 2.2.0.
  • Correctly set .cargo/config.toml variables for LIBTORCH and LD_LIBRARY_PATH
  • Installed tts on the system. It can use a different speech synthesizer by tweaking src/processing/speech_synthesizer.rs

Directory structure

|- models/              (WhisperAI models for speech recognition)
|- config/              (Supported command map and other configuration options)
|- src
    |- core/            (Shared functionality code and reusable components.)
    |- processing/      (Main processing pipeline for jarvis functionality)
    |- model/           (Model files)
    |- errors/          (Jarvis error handling)
    |- traits/          (Shared trait definitions)

Supported commands

To define commans that jarvis can run there's an example command_map.yaml present in config. It has the following structure:

commands:
    - "<LOCATION>":
        - <ACTION>:
            - <SUBJECT>
            - ...
            - <SUBJECT>
        - <ACTION>:
            - <SUBJECT>
            - ...
        - more actions...
    - more locations...

A concrete example

commands:
  - "living room":
    - switch:
      - light
    - gradient:
      - windowblinds
      - temperature
  - "dining room":
    - switch:
      - light

Locations are recognized automatically and can be arbitrary strings.

Actions are used to define what operation can be performed. Currently supported actions are:

  • switch used for toggling the subject on or off.
  • gradient used for a continuous variable that can be raised or lowered. Also supports turning all the way up or down eg. "Shut the window blinds".
    To define additional actions src/model/command_action.rs should be modified.

Subjects are things that actions can be performed upon. Currently supported are: light, teapot, window blinds, temperature, ventilator. They are easy to add and can be added in src/model/command_subject.rs

Running Jarvis

To run you can use cargo run or cargo run --release for the optimized version. Jarvis is a multithreaded app and shutting it down may have a short delay while all auxiliary threads wind down. Depending on the models you're using it can take up a significant amount of memory to run (around 1GB).

Whisper models

Additional language recognition models can be downloaded from https://ggml.ggerganov.com/. I recommend placing them in models and you can also tweak the default_model_path in speech_recognizer.rs.

License

Copyright (c) Jan Halozan

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

Acknowledgements

This project uses the following open-source libraries:

  • rust-bert by Guillaume Bertaud, licensed under the Apache License 2.0.