Autonomous Driving Agent with KNN Classifier 🏎️

Overview 📖

This project was developed as part of the Artificial Intelligence: Methods and Applications (2023–2024) course and presented at the Granpremio MIVIA 2024 Final Contest.

The goal of the contest was to design and implement an autonomous driving agent capable of completing a lap in the shortest possible time using machine learning techniques.
Our solution adopts a K-Nearest Neighbors (KNN) classifier integrated into the TORCS racing simulator through a custom Java client.

The repository includes:

• code/ – full Java source code of the TORCS client implementing the KNN algorithm
• demo_videos/ – demonstration videos showcasing the autonomous driving performance
• report.pdf – a detailed report (in Italian) describing design choices, dataset preparation, experiments, and results

---

System Description ⚙️

The Java client communicates with the TORCS server via UDP protocol and manages driving based on configurable parameters.
The approach follows behavioral cloning, where the agent learns from human driving demonstrations using sensor–action pairs collected in a dataset.

---

Usage 🖥️

1. Open the terminal and the Torcs application.

2. Move into the src folder where the client source code is located and compile it using:

   javac -d ../classes src/*.java

3. After compilation, start a new race from the Torcs application, ensuring that the race player is configured as scr_server 1.

4. Once the race has started (Torcs will wait for a client connection), move into the classes folder from the terminal and run the client with:

   java scr.Client scr.SimpleDriver host:localhost port:3001 verbose on

5. From the terminal, choose the execution mode:

   • Manual Driving
     Enter the character m and press Enter.
     Use w to accelerate, a to steer left, d to steer right, and s to brake or reverse.
     To start recording race data into dataset.csv, press q in the JFrame that appears on screen.
     To stop recording, press e.

   • Autonomous Driving
     Enter the character a and press Enter.
     The car will drive autonomously using the data in dataset.csv.

---

Dataset 📊

If you want to replace the dataset with an existing one for testing:

1. Go to the classes folder.
2. Delete the existing dataset.csv.
3. Place your new CSV dataset in the same folder and rename it to dataset.csv.

The dataset must follow this format:

"ANGLE_SENS(Double);GEAR_SENS(Int);RPM_SENS(Double);SPEEDX_SENS(Double);SPEEDY_SENS(Double);SPEEDZ_SENS(Double);TRACK_SENS(Double[]);TRACK_POS_SENS(Double);ACCEL_ACT(Double);BRAKE_ACT(Double);CLUTCH_ACT(Double);GEAR_ACT(Int);STEERING_ACT(Double)"

---

Recommendations for Testing 🧪

• Use a plugged-in computer to ensure maximum performance and avoid slowdowns.
• Close unnecessary background applications (e.g., browsers), as they may reduce the performance of the KNN algorithm and affect lap completion times.

---

Deliverables 📦

According to the project requirements, the submission includes:

• Commented source code (code/)
• Demonstration videos (demo_videos/)
• Written report (report.pdf, in Italian) detailing:
  • Classifier design choices
  • Experiments conducted
  • Final evaluation of the KNN model