Week 3 Robotics Update: PID’s, Computer Vision, and Delving into Reinforcement Learning.

Greetings, fellow robotics enthusiasts! Welcome to Week 3 of my exciting robotics journey. This week has been filled with new challenges and fascinating discoveries. Let’s dive right into the tasks I have accomplished and the plans for the upcoming week.

License Plate Detector using Haar Cascade

One of the tasks I tackled this week was building a license plate detector using Haar cascades for car images. The goal was to learn basic detection.

Challenges:

• Fine-Tuning Parameters: Tuning the parameters of the Haar cascade for accurate license plate detection demanded careful experimentation to find the right balance between precision and recall. However, with the help of YouTube videos and documentation, I managed to overcome these challenges.

Wins:

• Successful License Plate Detection: After persistent efforts and fine-tuning, I achieved a satisfactory license plate detector. Witnessing the model accurately identify license plates in various car images was a rewarding achievement.

While I am aware that Haar Cascades is a viable approach, it might not be the most accurate and efficient method for license plate detection in all scenarios. Which is why I have a small spark to explore YOLO which is a deep-learning based detection technique next for future references.

Who knows? Maybe I might revisit this project and try it a different way.

Exploring Reinforcement Learning (RL)

I started a short applied course on Reinforcement Learning, driven by my interest in intelligent agents capable of making informed decisions in dynamic environments, particularly in the context of self-driving cars.To begin with, I tackled the classic RL problem of balancing a pole on a cart This problem is often used as a foundation to ease into the vast ocean of RL.

Challenges:

• Understanding RL Concepts: Reinforcement Learning has its complexities, and grasping the underlying concepts was a tall mountain to climb. Although I haven’t fully scaled it yet, I’m determined to keep on learning and improving.
• Algorithm and Policy Selection: With numerous RL algorithms and policies to choose from, deciding which one to use for my pole balancing problem proved to be challenging. After seeking guidance, I opted for the PPO (Proximal Policy Optimization) algorithm with an MLP (Multi-Layer Perceptron) Policy.

Exploring PID Controllers

As part of my self-study, I also delved into PID controllers and their applicability in robotics. Kajal Gada’s blogs have been invaluable in guiding my exploration. Understanding control theory and how PID controllers maintain stability in robotic systems has been a fascinating learning experience.

Balancing Learning Load

In my previous articles, I set a goal of completing a problem a day on Leetcode. However, this week, I couldn’t dedicate as much time to Leetcode questions as I would have liked. Realizing that I was overloading my brain with continuous streams of information, I consciously decided to ease my study pattern to avoid feeling overwhelmed.

After all Rome was not built in a day!

Next Week’s Plan

1)Creating a RL Model for a 2D Self-Driving Car

In the upcoming week, I plan to embark on my ambitious project of developing a reinforcement learning model for a self-driving car in a 2D racetrack. This project will put my RL skills to the test and pave the way for more advanced robotics applications.

2)Realigning with Algorithmic Questioning

I will get back on track with my algorithmic problem-solving practice on Leetcode and aim to tackle more challenging problems.

3)Implementing an Obstacle Avoidance Robot in Webots

Building on my knowledge from the line follower robot, I intend to create a robot capable of intelligent obstacle avoidance in the Webots simulation environment.

4)Advancing in PID Controllers and Inverted Pendulum Problem

I will continue my study of PID controllers and work towards applying them to solve the inverted pendulum on a cart problem.

I am excited for the challenges and triumphs that await me in Week 4 of my robotics journey. Stay curious, stay inspired, and let’s continue exploring the boundless possibilities of robotics together! Most importantly — be kind to yourself. Until next time!

Note:

If you’re as enthusiastic about robotics as I am and want to discuss anything related to robotics or other subfields, feel free to reach out on my LinkedIn. Let’s connect and have meaningful conversations about our shared interests. I’m always open to learning and growing together in this fascinating field. Looking forward to connecting with fellow enthusiasts like you!