https://github.com/mozturan/human-like_autonomousdrive3d_drl

Human-Like Autonomous Driving W/ Deep Reinforcement Learning in 3D environment
https://github.com/mozturan/human-like_autonomousdrive3d_drl

autonomous-driving deep-learning deep-reinforcement-learning donkey-car donkeycar simulation soft-actor-critic

Last synced: 7 months ago
JSON representation

Human-Like Autonomous Driving W/ Deep Reinforcement Learning in 3D environment

• Host: GitHub
• URL: https://github.com/mozturan/human-like_autonomousdrive3d_drl
• Owner: mozturan
• Created: 2024-04-04T12:26:37.000Z (over 1 year ago)
• Default Branch: main
• Last Pushed: 2025-01-18T18:03:42.000Z (9 months ago)
• Last Synced: 2025-03-21T05:43:03.053Z (7 months ago)
• Topics: autonomous-driving, deep-learning, deep-reinforcement-learning, donkey-car, donkeycar, simulation, soft-actor-critic
• Language: Python
• Homepage:
• Size: 610 KB
• Stars: 2
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# Human-Like Autonomous Driving W/ Deep Reinforcement Learning

The objective is to develop human-like driving behaviors in autonomous vehicles using a hierarchical deep reinforcement learning (DRL) framework. Visual feature extraction and dimensionality reduction from RGB camera images are conducted using an autoencoder, which allows the agent to analyze the driving environment more effectively. Soft Actor-Critic (SAC) algorithm is selected as the DRL method for decision-making and control in autonomous driving; additionally, action noise is incorporated to facilitate a more efficient learning process. Action filtering techniques, such as Exponential Moving Average (EMA) and Weighted Moving Average (WMA), are employed to smooth the movements of the steering and throttle, thereby enhancing driving fluidity.

**PS: Working on obstacle avoidence, currently.**

**The experiments carried out in this study aim to complete the course with a smooth human-like driving experience by making decisions and providing control with the proposed system, staying within the lane and at the targeted speed. The simulation environment required for the experiments was provided using the simulator of the Donkeycar project, an open-source autonomous vehicle driving platform.**

Some detailed information about the study is given below. The study is in the development process and will be edited in the near future.

Video to observe performances: [Video](https://www.youtube.com/watch?v=UJ_SdjIPPi8) 

____________________________________________

 | System Architecture For Training

 |:-------------------------:

 | ![Screenshot from 2024-10-07 12-44-58](https://github.com/user-attachments/assets/25889082-82de-4fc2-b3d4-198d7fdeee37)

Model Type | Description

:-------------------------:|:-------------------------:

SAC|Basic SAC (Reference)

NSAC| Action Noise Added SAC

NSAC-EMA| Noised + Action filtered SAC w/ Exponential moving average (EMA)

NSAC-WMA| Noised + Action filtered SAC w/ Weighted moving average (WMA)