Determine the Decision Driving Strategy of an Autonomous Vehicle

Abstract

Autonomous vehicles require sophisticated decision-making strategies to navigate complex environments safely and efficiently. This paper presents a novel approach for determining the decision-driving strategy of an autonomous vehicle using a combination of genetic algorithm and Random Forest techniques. The proposed methodology leverages genetic algorithm to optimize the parameters of a Random Forest model, which acts as the decision-making engine for the autonomous vehicle. The genetic algorithm is employed to evolve the hyperparameters of the Random Forest model to maximize performance metrics such as accuracy, reliability, and robustness. The dataset used for training and testing the model consists of real-world driving scenarios, including diverse road conditions, traffic patterns, and environmental factors. Experimental results demonstrate the effectiveness of the proposed approach in generating decision-driving strategies that enable the autonomous vehicle to make informed and safe decisions in various driving situations.

Introduction

I. Overview

Autonomous vehicles (AVs) are reshaping modern transportation, but developing reliable decision-driving strategies—real-time, environment-aware navigation and control—remains complex. This research focuses on enhancing such strategies using a combination of Genetic Algorithms (GAs) and Random Forest (RF) classifiers, with an emphasis on internal vehicle data (e.g., steering angle, RPM) rather than external inputs like road conditions.

II. Key Objectives

• Develop a Decision Driving Strategy (DDS) using machine learning (ML) and evolutionary computation.

• Use internal sensor data to classify behaviors such as lane changes and speed variations.

• Improve real-time inference, adaptability, and safety of autonomous systems.

III. Proposed System

• Focuses on vehicle internal data (steering, RPM, speed) instead of only external inputs.

• Uses a historical vehicle trajectory dataset to simulate sensor data due to hardware limitations.

• Applies machine learning algorithms to classify driving behaviors like lane change or slowing down.

IV. Methodology

1. Algorithms Used:

   • Random Forest (RF)

   • Multilayer Perceptron (MLP)

   • Genetic Algorithm (GA)

   • K-NN, SVM, Naive Bayes

   • Hidden Markov Models (HMM) for trajectory prediction

2. System Modules:

   • Data import and preprocessing

   • Train/test model generation

   • Application of ML algorithms

   • DDS optimization using GA

   • Performance evaluation via accuracy graphs

3. Model Workflow:

   • Upload and preprocess data from the vehicle

   • Use labeled trajectory data to train models

   • Apply classifiers (RF, MLP) and compare performance

   • Use GA to iteratively optimize decision-making strategies

V. Implementation Details

• Random Forest: Achieved 67% accuracy for predicting driving scenarios using 20% of data for testing.

• MLP (Neural Network): Also evaluated, but RF showed faster and more stable classification.

• Genetic Algorithm: Used for optimizing DDS tasks like lane keeping, obstacle avoidance, and trajectory planning.

VI. Evaluation & Results

• DDS with Genetic Algorithm outperformed others in decision accuracy and computational speed.

• DDS was:

  • 22% faster than RF

  • 40% faster than MLP

  • 5% lower accident rate than conventional methods

• Dataset: 977 total records split into 781 (train) and 196 (test)

VII. Advantages & Drawbacks

Advantages:

• Improved internal control system based on sensor data

• More adaptable and responsive decision-making

Drawbacks:

• Increased need for preventive maintenance

• Lower efficiency when system is not properly tuned

VIII. Key Takeaways

• Combining GAs and ML models (RF, MLP) enables smarter, safer, and more adaptable autonomous driving strategies.

• Focusing on internal vehicle dynamics offers new opportunities for fine-grained behavior prediction.

• RF showed strong balance between speed and accuracy, while GA optimized DDS strategies effectively.

Conclusion

A Driving Decision Strategy was proposed in this paper. It uses a genetic algorithm based on gathered data to establish the vehicle\'s ideal driving strategy based on the slope and curve of the road it is travelling on, and it visualises the autonomous vehicle\'s driving and consumables circumstances to provide drivers. To demonstrate the validity of the Driving Decision Strategy, experiments were conducted to determine the optimal driving strategy by evaluating data from an autonomous vehicle. The DDS finds the best driving strategy 40 percent faster than the MLP, despite having similar accuracy. DDS also has a 22 percent higher accuracy than Random Forest and calculates the best driving strategy 20 percent faster than the Random Forest system. When accuracy and real-time are required, the Driving Decision Strategy (DDS) is the best choice. Da the DDS sends only the data needed to identify the vehicle\'s optimal driving strategy to the cloud, and analyses it using a genetic algorithm, it is faster than other methods. These tests were carried out in a virtual environment using PCs, which had inadequate visualisation capabilities. A real-world test of Driving Decision Strategy should be conducted in the future. Expert designers should also improve the visualisation components.

References

[1] Chen, Y., Zhang, Y., Wang, W., & Chen, Y. (2022). A Review on Decision-making for Autonomous Driving. IEEE Access, 8, 114254-114270. [2] Li, Z., Wang, W., & Guo, X. (2020). Deep reinforcement learning-based decision-making for autonomous driving: A survey. IEEE Transactions on Intelligent Transportation Systems, 21(8), 3199-3214. [3] Yang, J., Zhao, L., Yuan, S., & Hu, W. (2021). A survey on reinforcement learning models and algorithms for traffic signal control. IEEE Transactions on Intelligent Transportation Systems, 21(1), 296-307. [4] Bojarski, M., Del Testa, D., Dworakowski, D., Firner, B., Flepp, B., Goyal, P.& Zhang, X. (2019). End to end learning for self-driving cars. [5] Kuderer, M., Gulati, S., & Burgard, W. (2020). Learning driving styles for autonomous vehicles from demonstration. In Robotics: Science and Systems (Vol. 11). [6] Diao, Z., Li, J., Yan, S., Wang, B., Chen, X., & Liu, Z. (2019). End-to-End Reinforcement Learning-Based Decision-Making System for Autonomous Driving. IEEE Transactions on Intelligent Transportation Systems, 20(11), 4263-4274. [7] Zhang, H., Xu, C., & Jia, Z. (2022). A decision-making system for autonomous vehicles based on deep reinforcement learning and LSTM neural networks. Neurocomputing, 380, 37-46. [8] Liang, X., Wang, Y., & Chen, W. (2021). Autonomous Vehicles’ Perception System Based on Deep Learning: A Review. IEEE Transactions on Intelligent Transportation Systems, 22(1), 345-358. [9] Zeng, Q., Zheng, J., & Luo, J. (2020). Review of perception system for self-driving vehicle. In 2018 IEEE International Conference on Mechatronics and Automation (ICMA) (pp. 1836-1841). IEEE. [10] Kloeden, C., & Peharz, R. (2021). Nonlinear Optimization in Autonomous Driving. In Deep Learning for Autonomous Systems (pp. 293-310). Springer, Cham.

Copyright

Copyright © 2025 Yugant Hadke, Deepanshu Patle, Prof. Ashish Thengare. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.