Fully Autonomous Car

Project Title

Fully Autonomous Car

Project Area of Specialization Artificial IntelligenceProject Summary

We present an approach towards mapping and safe navigation in real, large-scale environments with an autonomous car. The goal is to enable the car to autonomously navigate on roads while avoiding obstacles and simultaneously learning an accurate three-dimensional model of the environment. In order to achieve fast lane mapping, our system compresses the sensor data using multi-level surface maps. The overall system runs on a modified car prototype equipped with different types of sensors (Encoders, Camera, IMU, GPS, Kinect). These sensors are fused to send input for multiple algorithms  that generate a trajectory which is achieved through steering angle and velocity contol.

Project Objectives

This project aimed to develop a prototype of single seat autonomous car which can eliminate the dependency of driver. The project consisted of a mechanical control system, electrical control system incorporated with scene segmentation and localization to drive autonomously. Using lane mapping and ORB SLAM, this vehicle is made autonomous which results in making fair and intelligent decisions while electrical control system (PID and MPC) drive the vehicle smoothly.Our aims for the project are as follows:

• To make the car drive-by-RC.
• To develop efficient autonomous system implemented through Computer Vision
• The car will follow a plain road which is properly marked.
• The traffic and obstacles must be minimal for the algorithm to work efficiently.

Project Implementation Method

Hardware Stack

1. 3 wheeled Akerman chassis, 300W Geared DC motor
2. Honda Life Power Steering
3. Nvidia Jetson Tx2 for executing trained model.
4. IBT2 40A H bridge, Custom Motor drive (IRF1407 MOSFETs)
5. High precision wheel encoders,
6. HD industrial camera
7. Kinect V2
8. Ethernet Hub
9. Pixhawk
10. Rheostat Breaking

Software Stack

1. Dead Reckoning using wheel encoders.
2. Fusing Encoders, Visual odometry using camera, IMU and GPS using navigation stack on ROS
3. Using a predefined map, cars and car’s position Hybrid A* can be used to generate possible trajectories.
4. Vehicle Detection using tiny YOLO on Nvidia TX2 running at 30Fps
5. Advanced Lane Detection using canny, sobel, perspective transform and poly-fitting detected lines to compute steering angles. Tested through Udacity Hardest challenge
6. Camera and IMU calibration using Kalibre
7. PID control of steering and drive motors
8. OSM Cartography to track vehicle position using GPS 
9. ORB SLAM and VINS MONO for visual odometry.

All processing was done on Nvidia Jetson TX2, Vehicle control through Arduino and Laptop was used as interface.

Communication was done with ethernet hub.

Benefits of the Project

1.2 Million People are killed on roads every year. It is same as the 747 falling out of the sky every year. This is only the loss of life the property damage due to these accidents are immeasurable. Traffic is getting worse, on an average a person spends 50 minutes traveling daily. Which is basically 162 lives wasted on roads daily in the US alone. There are also people who can’t drive due to numerous reasons. Using intelligent autonomous system in the car, ease for the life of people is created. This is research oriented project since this project is combination of electrical, mechanical and autonomous systems. Since this is electric car, so eliminates the threatening environment problems caused by the conventional cars due to their fuel engines. Everyday a lot accident takes place and people lost their precious lives. This car will also cater these accidents and will reduce their number as it is based on highly refined technology of end to end machine learning which have less chances of error. Besides, speed control of car and its position on the lane can also be more easily maintained using its autonomous features. Though the work on this project is going on in the world but in Pakistan, this is first time for any group to work on it. It is an innovative idea which grabs the attention of the customers and other people.

This autonomous system can be implemented for:

• Cargo
• Delivery
• Defence
• Shuttle service
• Car sharing

Technical Details of Final Deliverable

Packages and tools used to implement Self driving car.

1. OpenCV
2. ROS Navigation and Localization Stack
3. Extended Kalman Filter
4. Hybrid A*
5. MPC
6. Kalibre
7. OSM Cartography
8. ORB SLAM
9. YOLO
10. Google Maps API
11. Linux
12. Semantic Segmentation and Monocular depth using deep learning (low fps)

Final Deliverable of the Project HW/SW integrated systemType of Industry IT , Transportation Technologies Artificial Intelligence(AI), RoboticsSustainable Development Goals Good Health and Well-Being for People, Industry, Innovation and Infrastructure, Sustainable Cities and Communities, Responsible Consumption and ProductionRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	78800
Camera	Equipment	1	15000	15000
Motor drivers	Equipment	2	3000	6000
Battery	Equipment	3	2000	6000
Encoders	Equipment	2	2500	5000
IMU	Equipment	1	900	900
GPS	Equipment	1	700	700
Steering Mechanism	Equipment	1	15000	15000
Mechanical Work	Miscellaneous	1	5000	5000
Arduino	Equipment	1	1500	1500
Chassis	Equipment	1	18500	18500
Battery Charger	Miscellaneous	1	3000	3000
Buck convertor	Equipment	2	500	1000
SS Relay	Miscellaneous	1	1000	1000
Rheostat Break	Equipment	1	200	200