DynamicRAISS-VehicleSafety

Please find the Technical Paper of this Product here- Technical Paper

A dynamic perception-driven framework that proactively protects Vulnerable Road Users (VRUs)-especially pedestrians integrating real-time object detection, intent estimation, risk scoring, and control simulation into a unified safety system in EVs & autonomous vehicles. Product under company--Leameng Solutions Technologies (LeSo). The following is the Technical Product report (Unpublished draft) of the project. (Yet to be published paper)Technical Paper

DRAISS-VRU: Dynamic Risk-Aware Intent Sensing & Simulation for Vulnerable Road Users

Affiliations:

• Indian Institute of Technology Indore
• Leameng Solution Technologies (LeSo)

---

📘 Overview

DRAISS-VRU is a dynamic perception-driven framework that proactively protects Vulnerable Road Users (VRUs)—especially pedestrians—by integrating:

• Real-time object detection
• Intent estimation
• Risk scoring
• Override control simulation
• Acoustic warning system

This project simulates a unified safety system specifically designed for Electric Vehicles (EVs) and Autonomous Vehicles (AVs) using MATLAB Simulink.

📄 This repository hosts the technical project files, including:

• MATLAB .slx simulation model
• Python code for risk estimation
• AVAS logic design
• Technical Report (Unpublished Draft)

---

🎯 Key Features

1. Perception Layer (YOLOv8-based)

• Real-time pedestrian detection from camera feed using YOLOv8.
• Supports bounding box extraction and zone classification.
• Configurable for integration with LiDAR and RADAR (future upgrade).

2. Intent Estimation

• Zone-based logic (crosswalk, curbside, approach angle).
• Movement tracking to determine if a pedestrian intends to cross.

3. Risk Scoring Engine

• Python-based model using XGBoost classifier.
• Calculates risk levels based on velocity, intent, and proximity.
• Integrated into Simulink via MATLAB Function Block using py.*.

4. Simulink-Based Simulation

• Simscape vehicle modeling for kinematic behavior.
• Real-time simulation of vehicle override based on risk.
• Full vehicle path behavior simulation using input from the risk model.

5. AVAS (Acoustic Vehicle Alerting System)

• Dynamic AVAS tone modulation (volume + frequency) based on:
  • Risk level
  • Pedestrian distance
  • Intent probability
• Based on AIS 138 standards for quiet EV pedestrian alerts.

---

System Architecture

Logic Blocks Preview

Simulink Model- Vehicle Dynamics AVAS (Acoustic Vehicle Alerting System ) Logic

Results

Pedestrian Detection & Intent Prediction

Testing Results

MATLAB results under 2 sec sample time simulation

---

Guide to Run

Requirements

• MATLAB + Simulink (R2022b or later recommended)
• Python 3.8+
• Python packages: joblib, xgboost, numpy
• YOLOv8 environment (Ultralytics) for external detection/training

Simulation Setup

• Clone this repository:

git clone https://github.com/kaustuv-d/DynamicRAISS-VehicleSafety.git
cd DRAISS-VRU

• Open the Simulink model: rwdEV_test01.slx.

• Configure Python in MATLAB (point to your Python interpreter that has xgboost):

pyenv('Version','<path_to_python_exe>')

• Place xgb_risk_model.pkl in your MATLAB working directory (or update the path in your code block).
• Confirm audio output device (for AVAS) if using the Audio Device Writer block.
• Run simulation: Press Ctrl + T (or click Run). Recommended 10–20 s test run.

📊 Python Risk Logic (Used in Simulink MATLAB Block)

This Python snippet is called from within a MATLAB Function block via Python integration (py.* interface):

import joblib
import numpy as np

# Load trained XGBoost model (multi-class: Safe, Alert, Emergency)
model = joblib.load('xgb_risk_model.pkl')

# x_input = [v, x, y, brake, steer, a, angle_to_ped, mu, vis, bdist]
risk_probs = model.predict_proba([x_input])[0]
risk_level = int(np.argmax(risk_probs))  # 0 = Safe, 1 = Alert, 2 = Emergency

📈 Results & Performance (Example Metrics)

• Real-time pedestrian detection at ~30 FPS on RTX 3070 (YOLOv8n runtime).
• Risk classification latency ~20 ms per feature update (Python XGBoost).
• AVAS tone response <50 ms from risk trigger (Simulink + audio pipeline).
• Override logic shows controlled deceleration in A2, full brake in A3 events.
• Replace these with your measured values once you log actual runs.

📌 Future Enhancements

• Integration with CARLA Simulator or ROS2 for hardware / HIL testing.
• LiDAR + RADAR fusion (Kalman / EKF) for robust distance + velocity estimation.
• Embedded deployment on Jetson Nano / Raspberry Pi for edge inference.
• Extend to other VRUs: cyclists, wheelchair users, and small mobility devices.
• Transformer-based deep intent prediction using pose + trajectory sequences.

📚 References

• AIS 138 / AIS 189 – Acoustic standards for EV pedestrian alerting (India).
• YOLOv8 Documentation
• XGBoost Documentation
• MATLAB Simulink Vehicle Dynamics Toolbox (MathWorks).
• THOR-AVAS / industry AVAS concepts for EV pedestrian safety.

Author

Kaustuv Devmishra

Final Year UG | Indian Institute of Technology Indore (IIT Indore) | Dept. of Mechanical Engineering

Vehicle Safety AI Intern – Leameng Solution Technologies (LeSo), Bangalore

📧 Email: kaustuv2003@gmail.com

🔗 LinkedIn: Kaustuv