Reinforcement Learning Based Autonomous Navigation in Dynamic Environments

With technological advancement, we are moving to future with autonomous systems which require little to no manhandling. In this project, our aim is to navigate between two locations without any human aid. Machine Learning can be used to navigate a ro

Project Title

Reinforcement Learning Based Autonomous Navigation in Dynamic Environments

Project Area of Specialization

Robotics

Project Summary

With technological advancement, we are moving to future with autonomous systems which require little to no manhandling. In this project, our aim is to navigate between two locations without any human aid. Machine Learning can be used to navigate a robot through a dynamic environment and learn to correct its navigation skills quickly. In this project, we are using reinforced learning. Robot will monitor the surroundings and provide this live feed to a processor. The processor will make the most benefitting decision based on reinforced learning algorithms. This decision making will help the robot in adapting a changing environment during its navigation.

Project Objectives

Prime objectives of this project are:

• To develop/modify a prototype of a robotic car that can work according to given instructions.
• Data collection of the surrounding environment.
• Develop an algorithm based on reinforcement learning that can differentiate and classify objects present in a typical environment.
• Application of the developed algorithm on the collected raw data into labeled form.
• Making the system capable of making decision according to the processed data.
• Integrating the robotic car and algorithm such that the car can navigate in a dynamic environment so that it can detect, avoid and circumvent an object.

Project Implementation Method

Our proposed system uses technique named Computer Vision for object detection. Computer Vision needs digital data to perform its functionality, so a couple of cameras will provide a live feed. A high processing controller (Raspberry Pi) will then work on the received data to sort out which object to avoid, when to change lane or its better to stop. Here a decision will be made which will then be forwarded to the prototype car for execution. The car will be fine with a fairly average processor (like Arduino or STM). Now for final step, car will act upon the instruction provided. At the end, our aforementioned prototype will perform task a normal human driver can perform on the road.

Benefits of the Project

• By using this approach, we can identify  trends and patterns easily.
• It provides us with automation of the various human tasks.
• It continuously improves itself overtime.
• Human related accidents will tend to decrease.

Technical Details of Final Deliverable

Software and hardware tools:

• Raspberry Pi
• Arduino
• Camera
• DC and servo motors
• Python
• Machine learning libraries of python
• Reinforcement Learning algorithms
• Color and object segregation filters

Final Deliverable of the Project

HW/SW integrated system

Core Industry

Transportation

Other Industries

Core Technology

Robotics

Other Technologies

Sustainable Development Goals

Industry, Innovation and Infrastructure

Required Resources

If you need this project, please contact me on contact@adikhanofficial.com