In this ever-changing world, everything is tending towards automation day by day and in this situation sensors, actuators and robot interaction with human plays an important role. In this particular project, we have designed a robot car that has light sensors to detect the light and follow it with the assistance of actuators(Motors).

In collaboration with Matteo Ginesi, Github

Proximal Policy Optimization (PPO) is a popular state-of-the-art Policy Gradient Method. It is supposed to learn relatively quickly and stable while being much simpler to tune, compared to other state-of-the-art approaches like TRPO, DDPG or A3C. This makes PPO often the first choice when it comes to solving a problem for the first time. PPO strongly builds on Trust Region Policy Optimization (TRPO).

Implementing Optimistic Adaptive Acceleration For Optimization also known as Optimistic-AMSGrad can be a challenge when working with large datasets such as CIFAR-100. To reduce the training time which would take a pretty long time and need additional computing resources. We are using CIFAR-10 datasets which as 10 different categories of images containing a fair amount of 600 images per category. As this is an extension for AMSGrad we have to compare the AMSGrad results with OPTIMISTIC-AMSGrad. Optimization algorithms are vastly used in various models such as Resnet18, Googlenet, Densenet, etc. In our implementation, we have used the Resnet18 model with Optimistic-AMSGrad and AMSGrad optimization algorithms which demonstrate that Optimistic-AMSGrad improves AMSGrad.

This is a short write-up and don’t use it to cheat in exams. If you do that you may find yourself in a position that you cannot go back and correct. Of course, this is meant only for educational purposes. And I only did it to satisfy my ego because people say that it is impenetrable(As I say nothing is 100% secure).