top of page

Automonous Driving Agent

By Team AgentSmart
屏幕截图 2021-04-20 090103.png

01

Introduction

Definition: A self-driving car is a vehicle that is capable of sensing its environment and moving safely with little or no human input.

​

Our team aims to build an intelligent agent that could be successfully driving by itself in a virtual game world. We will train our model based on  an open source  car-racing simulator on Github.

02

Game Features

We add new features to the car-racing simulator to build our training environment . 


Features Added: 

  • Obstacles on the road

  • Accelerator to give higher speed

  • Stars need to be collected

  • Rich terrain (e.g. ramp)

  • 5 sensors to detect obstacles

  • 5 sensors to detect stars and accelerators

  • Dense checkpoints to guide the car

Methodology

We try different methods to train our 

​driving agent in order to make it smarter。

03

  • Reinforcement Learning
  • Multiple agent training
  • Imitation Learning
  • Convolutional Neural Networks
key features
Demo

Performance Description

The success rate of completing the whole circle without hitting walls is about 5%.
​
The average time cost to complete the whole circle is round 50s.
​
Compared to the human performance of 46 seconds on average, the agent’s performance is close to human.
​
​​
​
​
                              More details
Technology
Contact

Demo

bottom of page