Autonomous Vehicle Using Deep Reinforcement Learning

Project Title

Autonomous Vehicle Using Deep Reinforcement Learning

Project Area of Specialization Artificial IntelligenceProject Summary

Our targeted system comes under the domain of image processing and machine learning which will train our agent to perform tasks. The system name is Autonomous Vehicle. We have seen that the vehicles are automated by the old technologies and takes time to get automated while our technology focuses on automating it more efficiently and quickly.

Working on both the software and hardware was the part of our interest so we decided to solve this problem. Our proposed idea will automate the vehicle much quicker using Reinforcement Learning technique which is a sub category of machine learning, where the old technologies failed to do that.

Project Objectives

Considering the problem “How efficiently we can automate the vehicle”, One of the main Technology we are going to use for the solution is “Deep Reinforcement Learning”. As compared to the previous technologies, where the automation of anything took place within months, we would be dealing with the creation of environment along with some of the component in it like states, actions, agents and reward. These components will help us training our model very quickly.
If we talk about this within Autonomous Vehicle Domain, where we will be considering Road as environment, moving forward as states, the vehicle as agent and the reward as how accurately the trained model is performing, this technology is going to automate our vehicle much quicker as compare to others

Project Implementation Method

As the proposed system is not dependent on the specific environment but here, we are going to mention the road detection scenario. Firstly, our vehicle is ready to be trained on the road. The job of this vehicle is to detect the left lane, right lane and the straight line. This is the first phase of the training. Once this training phase is completed, we will move to the object detection. Training of vehicle in such a way that once it goes off the road, we will give negative reward to our system (agent in system) during the training depending upon the timestamps.After the complete training, the vehicle will be approximately 90% accurate to detect the road and objects and work accordingly

Benefits of the Project

• This system will use strong algorithm to determine appropriate stopping distance, distance from?? another?? vehicle?? and?? other?? data ??that?? decreases ??the ??chance ??of?? accidents
• This system only focuses on the environment and to take decisions accordingly, so it is an ideal motorist. It would take a lot of danger out of the equation entirely.
• This system will be much more demanding as compare to others because the technology is building the system more efficiently.    
• The same model of automation can be applied in multiple domains.
• Automating the vehicle is much more efficient with this proposed technology as compared to the previous technologies in terms of cost.
• Other than automating the vehicle quickly, that vehicle will perform the tasks within specified time according to his capability.

Technical Details of Final Deliverable

We will use procedural approach and we will develop each module separately, test them and integrate them.

We are using OpenCV and Python for image processing and deep learning algorithms, as both these languages are procedural languages

Reinforcement learning will help us to come up with the best possible actions on each input during the working process on the backend. This will also help us in coming up with the best possible action by making the Reward matrix table

Image Processing is the process which will help us in getting the images (frames/states) during the runtime. These frames are going to act as a input to our Neural Network during the training

Final Deliverable of the Project HW/SW integrated systemType of Industry IT Technologies Artificial Intelligence(AI)Sustainable Development Goals Industry, Innovation and InfrastructureRequired Resources