پروژه های پارسکدرز

Adaptive Cruise Control Using Deep Reinforcement Learning: Analysis of V2V Communication Delay Effects in Single-Vehicle and Platoon Scenarios
🔸 پروژه ویژه

#matlab, #automotive, #english_(us)

This project focuses on the design and analysis of an Adaptive Cruise Control (ACC) system with a particular emphasis on energy savings. The work is structured in two main stagesIn the first stage, the control strategy is evaluated on a single ego vehicle following a leading vehicleIn the second stage, the study is extended to a vehicle platoon, enabling the assessment of Cooperative Adaptive Cruise Control (CACC) behaviour
The control strategy adopted in this thesis is based on Deep Reinforcement Learning (DRL)
When extending the problem from a single follower vehicle to a full platoon, the system becomes cooperative, meaning that vehicles exchange information via Vehicle-to-Vehicle (V2V) communication.
In this project, we initially assume a perfect V2V communication channel, as is commonly done in many baseline studies.However, representing communication delay is essential for realistic evaluation.
Therefore, a communication delay component is introduced into the V2V model, and its effect is tested on:
The existing DRL agent, which was originally trained without delayA new or retrained DRL agent, potentially designed to be delay-awareThis allows evaluating:
how sensitive the RL controller is to communication latencyhow delay degrades string stability, energy efficiency, and comfortwhether retraining under delayed conditions improves robustness
In platoon control with artificial intelligence, two architectural options exist:
Option 1 — Centralized AI controllerA single agent controls the entire platoon.
This agent could theoretically be deployed in the cloud or infrastructure.Vehicles simply execute commands from the central controller.Option 2 — Distributed multi-agent control (our approach)Each vehicle carries its own RL agent and makes decisions locally.
The distributed design is more scalable and realistic for autonomous vehicles.In this context, modern Multi-Agent Reinforcement Learning (MARL) algorithms allow multiple vehicles to be trained simultaneously in a shared environment, where they learn:
how to maintain safe distanceshow to coordinate accelerationhow to coexist and minimize overall energy consumptionIn our scenario with a platoon of six vehicles, the agents can learn cooperative behaviour and inter-vehicle interactions through MARL paradigms.
این پروژه بر طراحی و تحلیل یک سامانه کروز کنترل تطبیقی (ACC) با تمرکز ویژه بر صرفه‌جویی در انرژی تمرکز دارد. کار در دو مرحله اصلی ساختاردهی شده است:
مرحله اول: ارزیابی استراتژی کنترل روی یک خودروی دنبال‌کننده (ego vehicle) که یک خودروی پیشرو را دنبال می‌کند.مرحله دوم: گسترش مطالعه به یک ستون خودرویی (platoon) برای بررسی رفتار کروز کنترل تطبیقی مشارکتی (CACC).
استراتژی کنترلی استفاده‌شده در این پایان‌نامه بر پایه یادگیری تقویتی عمیق (DRL) است.
با گسترش مسئله از یک خودروی دنبال‌کننده به کل یک ستون خودرو، سیستم حالت مشارکتی پیدا می‌کند؛ به این معنا که خودروها از طریق ارتباط بین‌خودرویی (V2V) با یکدیگر تبادل اطلاعات دارند.
در این پروژه، در ابتدا فرض می‌شود که کانال ارتباطی V2V بدون نقص است؛ همان‌ط...

توسط خریدار: کاربر739557

اطلاعات بیشتر:
https://parscoders.com/project/588951/adaptive-cruise-control-using-deep-reinforcement-learning:-analysis-of-v2v-communication-delay-effects-in-single-vehicle-and-platoon-scenarios

t.me/ParscodersCom

113 views00:23