1 of 12

Self-Driving Vehicular Project

Team Members:

Adarsh Narayanan, Joshua Menezes, Christopher Sawicki, Tommy Chu, Brandon Cheng, Ruben Alias, Aleicia Zhu, Ranvith Adulla, Arya Chhabra, Suhani Sengupta

This work was supported in part by the NSF REU program and the donation from nVERSES CAPITAL

2 of 12

Introducing the Team

Tommy Chu

Rutgers ECE UG

Adarsh Narayanan Rutgers ECE UG

Joshua Menezes

Rutgers ECE UG

Christopher Sawicki

Cornell MechE UG

Brandon Cheng

Rutgers ECE UG

Ruben Alias

Rutgers ECE UG

Aleicia Zhu

High School

Ranvith Adulla

High School

Suhani Sengupta

High School

Arya Chhabra

High School

​

3 of 12

Goal:

Build two model cars that are able to drive through the miniature smart city

​

RASCAL

Robotic Autonomous Scale Car for Adaptive Learning

SCAMP

Self-guided Computer Assisted

Mecanum Pathfinder

​

  • Restricted to real car movement
  • Use machine learning to drive autonomously

​

  • Mimic a real car’s path
  • Simulate traffic for autonomous car

4 of 12

RASCAL

(Robotic Autonomous Scale Car for Adaptive Learning)

  • Ackerman Steering
  • Differential gear system
  • Imitate a real car’s motion

5 of 12

Software Architecture

Pure Pursuit Feedback Loop

ML Model Inference Path

6 of 12

Web Display

  • Provides visual interface
  • ROS node runs Flask server
  • Add commands and points from any ROS node

e.g. city outline,

commands for car parameters (speed & pos),

editing paths/pure-pursuit

​

Intersection

7 of 12

Pure Pursuit Results:

  • Goal: Consistently follow a path
  • Path following algorithm
    • Create path on web display
    • Car self-adjusts to stay on path
  • Odometry to track car position

8 of 12

Machine Learning

  • Goal: drive using just camera feed
  • Does not require pre-programmed path
  • Learns from example data collected by pure pursuit
  • Pytorch
    • Imitation Learning Model
    • CNN (Convolutional Neural Network)

9 of 12

SCAMP

(Self-guided Computer Assisted Mecanum Pathfinder)

  • Path following
  • Mecanum wheels
  • Documentation to replicate

Car Front

10 of 12

SCAMP: Data Flow

relative (x, y, 𝜃)

𝜃

goal (x, y, 𝜃)

Motor power

+ motor direction

world (x, y)

world (x, y)

goal (x, y)

Inertial Mass Unit (IMU)

Teensy 4.0

Microcontroller

Motor Driver

Jetson Nano

Computer

Web server

Flask Server

Serial

Raw

motor

signal

Motor + Encoder

11 of 12

Future

RASCAL

  • Simultaneous localization and mapping (SLAM)
  • Integration with intersection cameras or larger field of view

SCAMP

  • Self orientation (SLAM/Intersection Cameras)
  • Path extraction from video
  • Instantaneous speed

​

12 of 12

Questions?