LEADS: Lightweight Embedded Assisted Driving System

LEADS is a lightweight embedded assisted driving system. It is designed to simplify the development of instrumentation, control, and analysis systems for racing cars. It is written in well-organized Python and C/C++ and has impressive performance.It is not only out-of-the-box but also fully customizable. It provides multiple abstract layers that allow users to pull out the components and rearrange them into a new project. You can either configure the existing executable modules (LEADS VeC) simply through a JSON file or write your own codes based on the framework as easily as building a LEGO.

The hardware components chosen for this project are geared towards amateur developers. It uses neither a CAN bus (although it is supported) nor any dedicated circuit board, but generic development kits such as Raspberry Pi and Arduino instead. However, as it is a high-level system running on a host computer, the software framework has the ability to adapt to any type of hardware component with extra effort.

This document will guide you through LEADS VeC. You will find a detailed version here.

🔗 Home

🔗 News

🔗 Docs

🔗 LEADS VeC Remote Analyst Online Dashboard

Why LEADS?

Rich ecology

LEADS

In a narrow sense, LEADS refers to the LEADS framework, which consists of an abstract skeleton and implementations for various businesses. The framework includes a context that interacts with the physical car, a device tree system, a communication system, a GUI system, and many specifically defined devices.

LEADS VeC

LEADS VeC composes a set of executable modules that are designed for the VeC (Villanova Electric Car) Project. These modules are customizable and configurable to a certain extent, yet they have limitations due to some assumptions of use cases. Compared to the LEADS framework, it sits closer to users and allows users to quickly test their ideas.

Accessories

Accessories contribute a huge portion to the richness of the LEADS ecosystem. These accessories include plugins that directly interact with the core programs, such as Project Thor, and standalone applications that communicate with LEADS through the data link, such as LEADS Jarvis.

As stated in its name, LEADS tries to keep a modular structure with a small granularity so that users have more flexibility in dependencies and keep the system lightweight. To do so, we split LEADS into multiple packages according to the programming language, platform, features, and dependencies, but generally, these packages are grouped into 3 types: LEADS (LEADS framework), LEADS VeC, and accessories.

Robust framework and good compatibility

The LEADS framework ensures that its applications, including LEADS VeC, have extremely high standards. They usually provide promising safety, but still, always keep our Safety Instructions in mind.

Most of the codes are written in Python and the dependencies are carefully chosen so that LEADS runs everywhere Python runs. In addition, on platforms like Arduino where we must use other programming languages, we try hard to keep consistency.

Modern GUI design

Dark mode support

LEADS inherently supports dark mode. You can simply change your system preference and LEADS will follow.

High performance

Even with extraordinary details and animation, we still manages to provide you with incredible performance.

Test Platform	Maximum Frame Rate (FPS)
Apple MacBook Pro (M3)	260
Orange Pi 5 Pro 8GB	200
Raspberry Pi 5 8GB	100
Raspberry Pi 4 Model B 8GB	60

External screens

LEADS supports an unlimited number of displays. You can enable a specific number of displays simply through a configuration.

Powerful ESC systems

DTCS (Dynamic Traction Control System)

DTCS helps you control the amount of rear wheel slip by detecting and comparing the speed difference between all wheels. It allows a certain amount of drift while ensuring grip.

ABS (Anti-lock Braking System)

ABS allows the driver to safely step on the brakes to the bottom without locking the brakes. Although this may increase braking distances, it ensures you always have the grip.

EBI (Emergency Braking Intervention)

EBI is available on most modern family cars. It actively applies the brakes at the limit of the braking distance. This system greatly reduces the probability of rear-end collisions.

ATBS (Automatic Trail Braking System)

ATBS monitors the steering angle and adjusts the brakes in time to ensure that the front of the car obtains the corresponding downforce. Its intervention will avoid under-steers (pushing the head) or over-steers (drifting).

Manual mode

It is not surprising that some racing events do not allow control of the vehicle beyond the driver. Therefore, we provide a manual mode where all assistance is forcibly disabled. You can simply set configuration manual_mode to True to enable this mode.

Plugins can be easily customized and installed in LEADS. It also comes with several existing ones including 4 ESC plugins that realize 4 ESC systems. All 4 systems have 4 calibrations: standard, aggressive, sport, and off. Their intervention comes later than the previous respectively.

Real-time data sharing and persistence platform

Thanks to our built-in TCP communication system, we are able to establish a powerful data link where the vehicle is the server and can connect to multiple clients. All data collected through LEADS can be distributed to the pit crew or anywhere far from the vehicle within an unnoticeable time, and that is not the end of the data's travel.

We are a big fan of data collection. With our data persistence technologies, all data are saved in popular formats such as CSV and processed in an intuitive way by LEADS VeC Data Processor.

During the development of LEADS, we have accumulated massive amounts of real-world data that can be made into public road datasets.

Live video streaming

As an organization that is deeply rooted in AI, visual data has special meaning to us. In LEADS, we value visual inputs more than any other framework. Our official support of low-latency video encoding and streaming has made LEADS natively suitable for AI applications.

LEADS is one of the few solutions on embedded single-board computers that support multiple cameras with an excellent performance.

Transparency between the driver and the pit crew

LEADS VeC Remote Analyst

LEADS VeC Remote Analyst is designed so that the pit crew can track the vehicle. It consists of a local web server and a frontend website.

With the LEADS, what the driver sees and feels is all shared with the pit crew in real time through our data link. The pit crew now masters every single bit of detail of the vehicle. This advantage will help the team avoid 99% of communication barriers.

Time machine with recorded data

Having the data saved by our data platform, you can easily run LEADS on a laptop and travel back to the time when the data was recorded.

AI-enhanced data analysis and driver training

Powered by rich datasets, our ambition is to change car racing as a whole, just as AlphaGo changed Go. This blueprint has never been such easy as today thanks to AI.

Citation

Should you find our work helpful to you, please cite our publication.

@misc{fu2024leadslightweightembeddedassisted,
    title = {LEADS: Lightweight Embedded Assisted Driving System},
    author = {Tianhao Fu and Querobin Mascarenhas and Andrew Forti},
    year = {2024},
    eprint = {2410.17554},
    archivePrefix = {arXiv},
    primaryClass = {cs.SE},
    url = {https://arxiv.org/abs/2410.17554},
}

Installation

Python

Note that LEADS requires Python >= 3.12. To set up the environment on Ubuntu by only a single line of command, see Environment Setup.

pip install "leads[standard]"

If you only want the framework, run the following.

pip install leads

This table lists all installation profiles.

Profile	Content	For	All Platforms
leads	Only the framework	LEADS Framework	✓
"leads[standard]"	The framework and necessary dependencies	LEADS Framework	✓
"leads[gpio]"	Everything "leads[standard]" has plug lgpio	LEADS Framework	✗
"leads[vec]"	Everything "leads[gpio]" has plus pynput	LEADS VeC	✗
"leads[vec-no-gpio]"	Everything "leads[standard]" has plus pynput	LEADS VeC (if you are not using any GPIO device)	✓
"leads[vec-rc]"	Everything "leads[standard]" has plus "fastapi[standard]	LEADS VeC Remote Analyst	✓
"leads[vec-dp]"	Everything "leads[standard]" has plus matplotlib and pyyaml	LEADS VeC Data Processor	✓

Arduino

You can install LEADS Arduino from Arduino Library Manager. Note that it is named "LEADS", not "LEADS-Arduino", in the index.

LEADS Framework

See Read the Docs for the documentation of how to customize and make use of the framework in your project.

Quick Start

Main

leads-vec run

Verify

leads-vec info

Replay

leads-vec replay

Replaying requires "main.csv" under the data directory. It as well accepts all optional arguments listed below.

Benchmark

leads-vec benchmark

Optional Arguments

Run the following to get a list of all the supported arguments.

leads-vec -h

Specify a Configuration File

leads-vec -c path/to/the/config/file.json run

You can use ":INTERNAL" to replace the path to leads_vec. For example, instead of typing ".../site-packages/leads_vec/devices_jarvis.py", simply use ":INTERNAL/devices_jarvis.py".

If not specified, all configurations will be default values.

To learn about the configuration file, read Configurations.

Specify a Devices Module

leads-vec -d path/to/the/devices.py run

You can use ":INTERNAL" to replace the path to leads_vec. For example, instead of typing ".../site-packages/leads_vec/devices_jarvis.py", simply use ":INTERNAL/devices_jarvis.py".

To learn about the devices module, read Devices Module.

Specify a Main Module

leads-vec -m path/to/the/main.py run

You can use ":INTERNAL" to replace the path to leads_vec. For example, instead of typing ".../site-packages/leads_vec/devices_jarvis.py", simply use ":INTERNAL/devices_jarvis.py".

Function main() must exist in the main module, otherwise an ImportError will be raised.

It needs to have the identical signature as the following.

def main() -> int:
    """
    :return: 0: success; 1: error
    """
    ...

Generate a Configuration File

leads-vec -r config run

This will generate a default "config.json" file under the current directory.

Register a Systemd Service

leads-vec -r systemd run

This will register a user Systemd service to start the program. The service script is usually located at "/usr/local/leads/venv/python3.12/site-packages/leads_vec/_bootloader/leads-vec.service.sh". Edit the file to customize the arguments.

To enable auto-start at boot, run the following.

systemctl --user daemon-reload
systemctl --user enable leads-vec

You will have to stop the service by this command otherwise it will automatically restart when it exits.

systemctl --user stop leads-vec

Use the following to disable the service.

systemctl --user disable leads-vec

Use Reverse Proxy

leads-vec -r reverse_proxy run

This will start the corresponding reverse proxy program as a subprocess in the background.

Customize the Splash Screen

leads-vec -r splash_screen run

This will replace the splash and lock screen with LEADS' logo.

Magnify Font Sizes

leads-vec -mfs 1.5 run

This will magnify all font sizes by 1.5.

Use Emulation

leads-vec --emu run

This will force the program to use emulation even if the environment is available.

Automatically Magnify Font Sizes

leads-vec --auto-mfs run

Similar to Magnify Font Sizes, but instead of manually deciding the factor, the program will automatically calculate the best factor to keep the original proportion as designed.

Remote Analyst

leads-vec-rc

Go to the online dashboard at https://leads-vec-rc.projectneura.org.

Optional Arguments

Run the following to get a list of all the supported arguments.

leads-vec-rc -h

Server Port

leads-vec-rc -p 80

If not specified, the port is 8000 by default.

Specify a Configuration File

leads-vec-rc -c path/to/the/config/file.json

If not specified, all configurations will be default values.

To learn about the configuration file, read Configurations.

Data Processor

leads-vec-dp path/to/the/workflow.yml

To learn more about workflows, read Workflows.

Environment Setup

This section helps you set up the identical environment we have for the VeC project. A more detailed guide of reproduction is available here. After the OS is set up, just run the one-line commands listed below. You may also choose to clone the repository or download the scripts from releases (only stable releases provide scripts).

These scripts currently only support apt as the package manager.

If you install Python using the scripts, you will not find python ..., python3 ..., pip ..., or pip3 ... working because you have to specify the Python interpreter such that python-leads ... and pip-leads ....

LEADS

You can simply run "setup.sh" and it will install everything including Python 3.12 all the optional dependencies of LEADS for you.

bash "setup.sh$(wget -O setup.sh https://raw.githubusercontent.com/ProjectNeura/LEADS/main/scripts/setup.sh)" && rm setup.sh || rm setup.sh

GPIOZero Compatibility

If you are using a GPIO board that is not a Raspberry Pi, you need to set the environment variable GPIOZERO_PIN_FACTORY to mock.

export GPIOZERO_PIN_FACTORY=mock

If you have registered the Systemd service, this line should be added to the service script as well like shown.

# adjust the arguments according to your needs
export GPIOZERO_PIN_FACTORY=mock
leads-vec -c /usr/local/leads/config.json run

Python

python-install.sh will only install Python 3.12 and Tcl/Tk.

bash "python-install.sh$(wget -O python-install.sh https://raw.githubusercontent.com/ProjectNeura/LEADS/main/scripts/python-install.sh)" && rm python-install.sh || rm python-install.sh

FRP

Install

We use FRP for reverse proxy. This is optional if you do not need public connections. If you want, install it through "frp-install.sh".

bash "frp-install.sh$(wget -O frp-install.sh https://raw.githubusercontent.com/ProjectNeura/LEADS/main/scripts/frp-install.sh)" && rm frp-install.sh || rm frp-install.sh

Configure

To configure FRP, use "frp-config.sh".

bash "frp-config.sh$(wget -O frp-config.sh https://raw.githubusercontent.com/ProjectNeura/LEADS/main/scripts/frp-config.sh)" && rm frp-config.sh || rm frp-config.sh

There are 4 arguments for this script, of which the first 2 are required.

bash "frp-config.sh$(...)" {frp_server_ip} {frp_token} {frp_port} {comm_port} && rm frp-config.sh || rm frp-config.sh

Uninstall

To uninstall LEADS, we provide an easy solution as well. However, it uninstalls a component only if it is installed through the way listed in Environment Setup.

bash "uninstall.sh$(wget -O uninstall.sh https://raw.githubusercontent.com/ProjectNeura/LEADS/main/scripts/uninstall.sh)" && rm uninstall.sh || rm uninstall.sh

Configurations

The configuration file is a JSON file that has the following settings. You can have an empty configuration file like the following as all the settings are optional.

{}

Note that a purely empty file could cause an error.

Index	Type	Usage	Used By	Default
w_debug_level	str	"DEBUG", "INFO", "WARN", "ERROR"	Main, Remote	"DEBUG"
data_seq_size	int	Buffer size of history data	Main	100
width	int	Window width	Main	720
height	int	Window height	Main	480
fullscreen	bool	True: auto maximize; False: window mode	Main	False
no_title_bar	bool	True: no title bar; False: default title bar	Main	False
theme	str	Path to the theme file	Main	""
theme_mode	bool	"system", "light", "dark"	Main	False
manual_mode	bool	True: hide control system; False: show control system	Main	False
refresh_rate	int	GUI frame rate	Main	30
m_ratio	float	Meter widget size ratio	Main	0.7
num_external_screens	int	Number of external screens used if possible	Main	0
font_size_small	int	Small font size	Main	14
font_size_medium	int	Medium font size	Main	28
font_size_large	int	Large font size	Main	42
font_size_x_large	int	Extra large font size	Main	56
comm_addr	str	Communication server address	Remote	"127.0.0.1"
comm_port	int	Port on which the communication system runs on	Main, Remote	16900
comm_stream	bool	True: enable streaming; False: disable streaming	Main	False
comm_stream_port	bool	Port on which the streaming system runs on	Main, Remote	16901
data_dir	str	Directory for the data recording system	Main, Remote	"data"
save_data	bool	True: save data; False: discard data	Remote	False

For device-related implicit configurations, please see the devices module.

Workflows

This only applies to LEADS VeC Data Processor. Please find a more detailed version here.

dataset: "data/main.csv"
inferences:
  repeat: 100  # default: 1
  enhanced: true  # default: false
  assume_initial_zeros: true  # default: false
  methods:
    - safe-speed
    - speed-by-acceleration
    - speed-by-mileage
    - speed-by-gps-ground-speed
    - speed-by-gps-position
    - forward-acceleration-by-speed
    - mileage-by-speed
    - mileage-by-gps-position
    - visual-data-realignment-by-latency

jobs:
  - name: Task 1
    uses: bake
  - name: Task 2
    uses: process
    with:
      lap_time_assertions: # default: []
        - 120  # lap 1 duration (seconds)
        - 180  # lap 2 duration (seconds)
      vehicle_hit_box: 5  # default: 3
      min_lap_time: 60  # default: 30 (seconds)
  - name: Draw Lap 5
    uses: draw-lap
    with:
      lap_index: 4  # default: -1
  - name: Suggest on Lap 5
    uses: suggest-on-lap
    with:
      lap_index: 4
  - name: Draw Comparison of Laps
    uses: draw-comparison-of-laps
    with:
      width: 0.5  # default: 0.3
  - name: Extract Video
    uses: extract-video
    with:
      file: rear-view.mp4  # destination to save the video
      tag: rear  # front, left, right, or rear
  - name: Save
    uses: save-as
    with:
      file: data/new.csv

Devices Module

Example

from leads import controller, MAIN_CONTROLLER
from leads_emulation import RandomController


@controller(MAIN_CONTROLLER)
class MainController(RandomController):
    pass

The devices module will be executed after configuration registration. Register your devices in this module using AOP paradigm. A more detailed explanation can be found here.

History Versions

These versions were used in practical races:

1. 0.9.5 (Villeneuve) - 2024-10-05 University of Waterloo EV Challenge

Collaborations

Community

Issues

Our team management completely relies on GitHub. Tasks are published and assigned as issues. You will be notified if you are assigned to certain tasks. However, you may also join other discussions for which you are not responsible.

Projects

You can have a look at the whole schedule of each project in a timeline using the projects feature.

Code Contributions

See CONTRIBUTING.md.