Determining whether a driver is operating a vehicle aggressively and pinpointing, for instance, which segment of a route they could drive more cautiously to save fuel, optimize vehicle resources, and reduce environmental impact is of interest to various individuals and businesses. Now, envision accomplishing this in real-time using Internet of Things (IoT) technology!
With this in mind, this repository contains an implementation of the Autocloud algorithm designed to run online on the Freematics ONE+. Clustering is performed using a soft sensor called the "radar chart area," essentially an indicator of vehicle resource utilization based on speed, throttle position, engine load, and rpm sensors. Once the area is calculated, the TEDA algorithm first checks if it is an outlier, and if not, it is grouped by Autocloud.
However, each driver exhibits a unique profile, not to mention that the same driver may drive differently depending on factors such as the route, traffic, weather, etc. This phenomenon, known as concept drift, needs to be addressed in real-time. In this regard, the concept of a dynamic window was introduced, where the latest outlier in a window becomes the center of the most typical cloud related to it, defining the window reset.
1 - Install Visual Studio Code
2 - Install PlatformIO (VSCode Extension)
3 - Clone this repository
git clone https://github.com/conect2ai/EAIS2024-Autocloud.git4 - Open freematics-autcloud/firmware_v5/telelogger project folder in PlatformIO, as shown in the figure below.
5 - Connect the Freematics One+ with the computer and power up it with the Freematics Emulator or in the vehicle.
6 - Compile, Upload and monitor serial (steps 1, 2 and 3, respectively in the figure below).
In the telelogger.ino file, you can customize the algorithms for your application. For example:
- To modify the reset windown lenght, you can change the value of this variable:
int max_outlier_window = 4;- To modify the TEDA's sensibility, you modify the initialization parameter:
TEDA teda(2.0);- To modify the sensors used in radar area calculation, you can modify the function and the struct below:
typedef struct {
float rpm;
float speed;
float throttle;
float engine;
} RADAR_DATA;
float calculate_area(RADAR_DATA& radar_data){
float rpm = radar_data.rpm / 100.0;
float speed = radar_data.speed;
float throttle = radar_data.throttle;
float engine = radar_data.engine;
float valuesNormalized[4] = {rpm, speed, throttle, engine};
float area = 0.5 * std::abs(
valuesNormalized[0] * valuesNormalized[1] +
valuesNormalized[1] * valuesNormalized[2] +
valuesNormalized[2] * valuesNormalized[3] +
valuesNormalized[3] * valuesNormalized[0]
) * std::sin(2 * M_PI / 4);
return area;
}- To modify the sensors and the times they are collected:
PID_POLLING_INFO obdData[]= {
{PID_SPEED, 1},
{PID_RPM, 1},
{PID_THROTTLE, 1},
{PID_ENGINE_LOAD, 1}
};All the information collected can be send to a server and can be stored locally in SD card. The hardware allows information to be sent both via Wi-Fi and 4G. You can configure things like Wi-Fi name and password, URL server, server port, protocol etc. in config.h file.
/**************************************
* Networking configurations
**************************************/
#ifndef ENABLE_WIFI
#define ENABLE_WIFI 0
// WiFi settings
#define WIFI_SSID "FREEMATICS"
#define WIFI_PASSWORD "PASSWORD"
#endif
#ifndef SERVER_HOST
// cellular network settings
#define CELL_APN ""
// Freematics Hub server settings
#define SERVER_HOST "hub.freematics.com"
#define SERVER_PROTOCOL PROTOCOL_UDP
#endif
// SIM card setting
#define SIM_CARD_PIN ""
// HTTPS settings
#define SERVER_METHOD PROTOCOL_METHOD_POST
#define SERVER_PATH "/hub/api"
#if !SERVER_PORT
#undef SERVER_PORT
#if SERVER_PROTOCOL == PROTOCOL_UDP
#define SERVER_PORT 8081
#elif SERVER_PROTOCOL == PROTOCOL_HTTP
#define SERVER_PORT 80
#elif SERVER_PROTOCOL == PROTOCOL_HTTPS
#define SERVER_PORT 443
#endif
#endif
// WiFi Mesh settings
#define WIFI_MESH_ID "123456"
#define WIFI_MESH_CHANNEL 13
// WiFi AP settings
#define WIFI_AP_SSID "TELELOGGER"
#define WIFI_AP_PASSWORD "PASSWORD"This section contains the graphs generated from the trips made by executing the code in this repository. The code that generated these graphs can be found in the notebook.ipynb file.
The initial trip involved a 2012 Honda Fit 1.6L with manual transmission in Natal, Rio Grande do Norte, Brazil. The driver began cautiously (blue points), shifted to a normal pattern (green points), and later drove more aggressively, noticeable by red points indicating higher engine load due to prolonged gear stretching. The algorithm accurately detected behavior changes, highlighted by green points amid red segments, often attributed to speed bumps and pedestrian crosswalks.
The second trip involved a 2014 Ford Fiesta 1.6L with automatic transmission in Natal, Rio Grande do Norte, Brazil. The driver exhibited cautious and normal driving for the most part. Occasionally, they accelerated abruptly to evaluate the algorithm's performance in limited aggressive data scenarios. The algorithm successfully captured changes in limited aggressive data and detected transitions between normal and cautious driving in segments, similar to observations in Driver 1's route.
Regarding the clustering of Driver 1, it is noticeable that a significant portion of the clusters overlapped, likely due to the prevalence of data from the normal and aggressive driving profiles. Despite the relatively separate cluster centers, the flow's shape influenced by the route's specificities resulted in closely positioned groups.
On the other hand, for the second driver, the notable distance of Cluster 1 in comparison to the other two clusters stands out, most likely attributed to abrupt accelerations performed by the driver within short time intervals.
Additionally, several cluster quality metrics were evaluated, providing insights into the effectiveness of the clustering process.
| Metric | Driver 1 | Driver 2 |
|---|---|---|
| Silhouette Score | 0.0137 | 0.1550 |
| Davies-Bouldin Score | 35.3533 | 9.2052 |
| Calinski-Harabasz Score | 22.9093 | 13.6478 |
| Dunn Index | 0.0000 | 0.0002 |
This project is licensed under the MIT License - see the LICENSE file for details.
The Conect2AI research group is composed of undergraduate and graduate students from the Federal University of Rio Grande do Norte (UFRN) and aims to apply Artificial Intelligence (AI) and Machine Learning in emerging fields. Our expertise includes Embedded Intelligence and IoT, optimizing resource management and energy efficiency, contributing to sustainable cities. In energy transition and mobility, we apply AI to optimize energy use in connected vehicles and promote more sustainable mobility.