search

ph4r05 / ph4-walkingpad

WalkingPad python controller
MIT License
72 stars 17 forks source link

readme

WalkingPad controller

Simple python script that can control KingSmith WalkingPad A1. Others report the similar models, such as R1 PRO work on the same principle.

The belt communicates via Bluetooth LE GATT. Only one device can be connected to the belt at a time, i.e., if original app is connected, the controller won't be able to connect.

Controller features

Demo

For the best understanding start jupyter-notebook and take a look at belt_control.ipynb

# Install jupyter-notebook
pip3 install jupyter

# Start jupyter-notebook in this repository
jupyter-notebook .

# open belt_control.ipynb

Library use-case

The main controller class is Controller in pad.py

Controller

Controller enables to control the belt via CLI shell.

Install the library:

pip install -U ph4-walkingpad

Start controller:

# Note: use module notation to run the script, no direct script invocation.
python -m ph4_walkingpad.main --stats 750 --json-file ~/walking.json

Or alternatively, if package was installed with pip:

ph4-walkingpad-ctl --stats 750 --json-file ~/walking.json

The command asks for periodic statistics fetching at 750 ms, storing records to ~/walking.json.

Output

---------------------------------------------------------------------------
    WalkingPad controller

---------------------------------------------------------------------------
$> help

Documented commands (use 'help -v' for verbose/'help <topic>' for details):
===========================================================================
alias      help     py  quit          set        speed   stop
ask_stats  history  Q   run_pyscript  shell      start   switch_mode
edit       macro    q   run_script    shortcuts  status  tasks

$> status
WalkingPadCurStatus(dist=0.0, time=0, steps=0, speed=0.0, state=5, mode=2, app_speed=0.06666666666666667, button=2, rest=0000)
$> start
$> speed 30
$> speed 15
$> status
WalkingPadCurStatus(dist=0.01, time=16, steps=18, speed=1.8, state=1, mode=1, app_speed=1.5, button=1, rest=0000)
$> status
WalkingPadCurStatus(dist=0.01, time=17, steps=20, speed=1.5, state=1, mode=1, app_speed=1.5, button=1, rest=0000)
$> speed 30
$> s
$> WalkingPadCurStatus(dist=0.98, time=670, steps=1195, speed=6.0, state=1, mode=1, app_speed=6.0, button=1, rest=0000), cal:  38.73, net:  30.89, total:  73.65, total net:  57.91
$> stop
$> start
$> speed 30
$> status

Due to nature of the BluetoothLE callbacks being executed on the main thread we cannot use readline to read from the console, so the shell CLI does not support auto-complete, ctrl-r, up-arrow for the last command, etc. Readline does not have async support at the moment.

OSX Troubleshooting

This project uses Bleak Bluetooth library. It was reported that OSX 12+ changed Bluetooth scanning logic, so it is not possible to connect to a device without scanning Bluetooth first. Moreover, it blocks for the whole timeout interval.

Thus, when using on OSX 12+:

Minimal required version of Bleak is 0.14.1

If the process is still crashing, it may be it does not have permissions to access Bluetooth. To fix it, add your Terminal app (in my case iTerm2.app) to System Preferences -> Security & Privacy -> Bluetooth.

Related resources: https://github.com/hbldh/bleak/issues/635, https://github.com/hbldh/bleak/pull/692

Profile

If the -p profile.json argument is passed, profile of the person is loaded from the file, so the controller can count burned calories. Units are in a metric system.

{
  "id": "user1",
  "male": true,
  "age": 25,
  "weight": 80,
  "height": 1.80,
  "token": "JWT-token",
  "did": "ff:ff:ff:ff:ff:ff",
  "email": "your-account@gmail.com",
  "password": "service-login-password",
  "password_md5": "or md5hash of password, hexcoded, to avoid plaintext password in config"
}

Note that once you use login command, other JWTs become invalid, e.g., on your phone. If you want to use the service on both devices, login with mobile phone while logging output with adb and capture JWT from logs (works only for Android phones).

Stats file

The following arguments enable data collection to a statistic file:

--stats 750 --json-file ~/walking.json

In order to guarantee file consistency the format is one JSON record per file, so it is easy to append to a file at any time without need to read and rewrite it with each update (helps to prevent a data loss in cause of a crash).

Example:

{"time": 554, "dist": 79, "steps": 977, "speed": 60, "app_speed": 180, "belt_state": 1, "controller_button": 0, "manual_mode": 1, "raw": "f8a2013c0100022a00004f0003d1b4000000e3fd", "rec_time": 1615644982.5917802, "pid": "ph4r05", "ccal": 23.343, "ccal_net": 18.616, "ccal_sum": 58.267, "ccal_net_sum": 45.644}
{"time": 554, "dist": 79, "steps": 978, "speed": 60, "app_speed": 180, "belt_state": 1, "controller_button": 0, "manual_mode": 1, "raw": "f8a2013c0100022a00004f0003d2b4000000e4fd", "rec_time": 1615644983.345463, "pid": "ph4r05", "ccal": 23.343, "ccal_net": 18.616, "ccal_sum": 58.267, "ccal_net_sum": 45.644}
{"time": 555, "dist": 79, "steps": 980, "speed": 60, "app_speed": 180, "belt_state": 1, "controller_button": 0, "manual_mode": 1, "raw": "f8a2013c0100022b00004f0003d4b4000000e7fd", "rec_time": 1615644984.0991402, "pid": "ph4r05", "ccal": 23.476, "ccal_net": 18.722, "ccal_sum": 58.4, "ccal_net_sum": 45.749}
{"time": 556, "dist": 79, "steps": 981, "speed": 60, "app_speed": 180, "belt_state": 1, "controller_button": 0, "manual_mode": 1, "raw": "f8a2013c0100022c00004f0003d5b4000000e9fd", "rec_time": 1615644984.864169, "pid": "ph4r05", "ccal": 23.608, "ccal_net": 18.828, "ccal_sum": 58.533, "ccal_net_sum": 45.855}
{"time": 557, "dist": 80, "steps": 982, "speed": 60, "app_speed": 180, "belt_state": 1, "controller_button": 0, "manual_mode": 1, "raw": "f8a2013c0100022d0000500003d6b4000000ecfd", "rec_time": 1615644985.606997, "pid": "ph4r05", "ccal": 23.741, "ccal_net": 18.933, "ccal_sum": 58.665, "ccal_net_sum": 45.961}

The benefit of having detailed data is an option to analyze data from the whole run, e.g., how step size varies over the time during one session, collect preferred speeds, etc...

Also, if the original app fails to fetch the final state from the Belt, having continuous data stream is helpful to avoid data loss.

Reversing Belt API

Easy way - Android logs

I used the easiest way I found - the original Android application is quite generously logging all Bluetooth requests and responses; and network requests and responses (JWT included).

After few trial/error attempts I managed to reverse binary packet protocol format. See pad.py for protocol internals.

You can query from the belt a status message (app does so each 750 ms, approx). The status contains speed, distance, steps, and very simple CRC code (sum of the payload). Interestingly, calories are not part of the status message and cannot be queried either.

For obtaining logs just plug Android phone via USB, enable development mode on the phone, enable ADB connection and run:

adb logcat

(Or use AndroidStudio)

You then can see the app communication with the belt in real-time. When using the app, it logs also requests so you can figure out how commands for e.g., speed change look like.

Medium - Bluetooth logs

Should vendor remove the logging from the app and you are unable to find APK in archives with the logging, you can always enable Bluetooth logs in the Phone development settings.

This approach is not that straightforward as from logs as you cannot see belt responses in real-time. The Bluetooth log can be obtained from the device via adb and opened in Wireshark.

You may need to do own journal with times and commands you issued so you can experiment with the belt (e.g., change speeds), the commands get logged to the Bluetooth log. Then after the experiment, download the Bluetooth log and map your log entries to the packets from the log.

This is substantially difficult compared to the easy way - message logs.

Hard way - Flutter disassembly

The original application is implemented in Flutter, so direct application reversing is quite painful process. Flutter compiles the source language (TypeScript I guess) to a binary form. It runs on top of a Flutter virtual machine, thus compiled binary has only one primary entry point, a dispatch function. Disassembly does not yield anything sensible, it requires special tools. Also, decompilation tools require the Flutter version to precisely match the version used to compile the application.

For those willing to spend time on this: 1, 2, 3, 4.

Hack way - BLE sniffer

Manual sniffer capture:

./nrf_sniffer_ble.sh --extcap-interface /dev/cu.usbserial-0001 --capture --fifo /tmp/fi

Alternatives

I was using the WalkingPad app to reverse engineer packet formats:

Other reported apps may be less obfuscated and easier to analyze (didn't test):

Protocol basics

Protocol internals are implemented in pad.py.

Example of a status message m:

f8a2010f01000fd10000ab0012ae3c0000003afd

When logged by the application, it is printed out as array if bytes:

[248, 162, 1, 15, 1, 0, 15, 209, 0, 0, 171, 0, 18, 174, 60, 0, 0, 0, 58, 253]

Meaning of some fields are not known (15) or the value space was not explored. m[15] could be for example heart rate for those models measuring it.

Related work

Another reverse engineer of the protocol (under GPL, tldr): https://github.com/DorianRudolph/QWalkingPad/blob/master/Protocol.h

Thanks

Thanks to all contributors and to the community.

This project was awarded by the Google Open Source Peer Bonus in Feb 2022.

Development

Install pre-commit hooks defined by .pre-commit-config.yaml

pip3 install -U pre-commit pytest mypy types-requests
mypy --install-types
pre-commit install

Auto fix

pre-commit run --all-files

Plugin version update

pre-commit autoupdate

Donate

Thanks for considering donation if you find this project useful:

Bitcoin

1DBr1tfuqv6xphg5rzNTPxqiUbqbRHrM2E

(No Lightning for now, hopefully soon)

Monero

87KDQUP7yVKd7inmX2WXuaQUBrxeGN9X9AuQwfaUkJ3KQXSRe6KbhnLRvWNK4mx2SeBwcFdHYgS71fzYFS5mtNf7Dn8SdpJ

PayPal

PayPal link