What is SlamJS

SlamJS is a free, opensource, easy-to-use and understand JavaScript library for camera pose estimation for VR, AR, XR use on mobile browsers. It is not yet in a production-ready state. However, the hope is that it provides a starting point for anyone wanting to experiment, play around with SLAM and its concepts.

Demo Video

https://github.com/mukulbasu/slamJS/assets/1706979/688b67ba-524e-4b4d-9edb-1957472d033f

Installation:

1. Make sure gcc, cmake, yarn are installed.
2. To build for Web, you have to set up emscripten. Follow the instructions from https://emscripten.org/docs/getting_started/downloads.html

Build and Running:

The application can be built and run locally against a test dataset or it can be run on the web browser. A sample dataset is provided in data/dataTest. The user can also choose to generate their own dataset for tests using the tools provided.

Building and running on Local machine:

This has been tested primarily on Ubuntu 22.04 LTS and Mac M2.

NOTE: Before building on Mac, open CMakeLists.txt and change set(LINUX TRUE) to set(LINUX FALSE). Also change set(APPLE_M FALSE) to set(APPLE_M TRUE).

Building for local machine:

1. Go to top level directory
2. yarn install
3. mkdir build; cd build
4. cmake ..
5. make -j10 Any time you update the libraries in the third_party folders, remember to rm -rf * everything in build folder and then redo from step 3 (cmake and so on).

Also the libraries to be copied to third_party/lib are all opencv and can be generated from https://github.com/mukulbasu/opencv. Follow the cpp build process mentioned in the README and copy over the libraries from the build/lib/* and build/3rdparty/lib/* into the third_party/lib/opencv/<arch> folder.

Running on local machine:

The executable binary is created in build directory as slamJS. A sample configuration file config/slamConfigMobile.js is provided. A test dataset data/dataTest is also present.

1. Go to top level directory
2. build/slamJS slamConfigMobile > debug/tmp
3. After this command completes running, move to Debugging.

Debugging:

The main debug website is built on React. The actual debug data on the website is served by a separate Node server (debug/server.js) that reads and serves data from debug/logs/debug.txt and debug/tmp. These files are generated once the command above is run.

1. Go to top level directory
2. cd debug
3. npm install; source startServer.sh
4. Open another terminal and in that type cd debug/react; npm install; npm start.
5. Open http://localhost:3000. Follow instructions on screen.

Building and running on Web:

This has been tested only on Android so far.

Building for Web:

1. Activate emscripten environment.
2. in the top level directory run: source build_wasm.sh

Running on Web:

Set up ADB wireless connection between your Android device and the local machine https://developer.android.com/tools/adb.

1. Go to top level directory
2. cd debug
3. source startServer.sh
4. Set up your android device for ADB connection.
5. On your local machine, open Google Chrome and point it to chrome://inspect/#devices. Make sure the port forwarding settings are set up to forward 8080 to localhost:8080.
6. On your Android Chrome browser open http://localhost:8080/slamDemo.html
7. Click Start AR to start the AR operation.
8. Stop AR is not tested yet. Reload the page instead to retest.

Building third party libraries

Pre-built binaries for third_party libraries are provided for linux and Mac ARM architecture. For other architectures, you might have build the binaries yourself as follows:

There are 2 main libraries in the folder : 1) Libraries from OpenCV 2) Libraries from Suitesparse

1. OpenCV: The libraries to be copied to third_party/lib are all opencv and can be generated from https://github.com/mukulbasu/opencv, branch: slamjs. Follow the cpp build process mentioned in the README and copy over the libraries from the build/lib/* and build/3rdparty/lib/* into the third_party/lib/opencv/<arch> folder.
2. Suitesparse: Clone https://github.com/DrTimothyAldenDavis/SuiteSparse. a. You might need to install openBLAS on linux using - sudo apt-get install libopenblas-dev b. In the top level directory, run the command: CMAKE_OPTIONS="-DBLA_VENDOR=OpenBLAS" make -j20 c. Go to build folder inside each of the respective directories like - CHOLMOD, AMD etc and copy the library from there to third_party/lib/suitesparse/<arch>/
3. G2O: The source code for the same is included in the third_party/src directory. Follow the instructions in README.md in third_party/src/go/ to build it for the required platform.

Running locally with your test data

Test data can be generated this way:

1. Go to top level directory
2. mkdir data
3. cd debug
4. source startServer.sh
5. On your mobile browser open - http://localhost:8080/test.html