Robotics and AI /Reinforcement learning

About the author

Sourabh Khemka

Radar Software Engineer at Steradian Semiconductors

Linkedin: https://www.linkedin.com/in/sourabh-khemka-b6894513a/

Weekend gossip! Let’s talk robotics

Marvelous inventions are always a fusion of multiple disciplines of engineering. When the best of all come together, a whole new dimension of research and development comes into existence.

One such beautiful and ingenious creation of mankind is Robotics.

Almost everyone having an YouTube account has seen videos of robots from Boston Dynamics or of Predator drones creating havoc in war-zones.

These state of the art technologies take shape when mechanical innovation bonds perfectly with the ones from electrical & electronics.

Let us imagine a very basic quadcopter, just like the one which you saw in Indian blockbuster 3-idiots ! Any layman can say that a humans uses a remote device to send signals to drones and drones rotate the motors to move :).

In reality, building a robust drone needs some serious kinematic models, dependable sensor units and a smooth electronic control. But wait ! If you’re into robotics, all you need is your passion to conquer it. Once you are into robotics, it is really tough to break-up with robots

In this post let’s focus not on building robots but on using robots. Robots run on Robot Operating System (ROS). ROS facilitates communication between sensors, actuators and the path planning and following algorithms.

A robot with a good ROS support and sturdy Mechatronics is a well built-up beast ready to roar. But what about its “brain” ? We talked about a robot’s body, its sensors, its operating system but who’s gonna take care of its movements ?

I guess you know humans are too lazy to control machines. If humans starts to manually control robots all day long then I should say Homo Sapiens are going back to ice age. How on earth can someone control let’s say a swarm of robots running here & there without Artificial Intelligence ! We use our brains to make robots and we have to let robots use their own to run.

AI and reinforcement learning in robotics

Robots have to use AI to classify what it gets from its sensors. Unmanned Vehicles also need to understand what the other idiot driving ahead of him is trying to do 😆. They use AI to predict dangers, to plan their path and to tune their control parameters so that naughty actuators don’t ditch them completely.

Reinforcement learning is the branch of machine learning that enables a robot to learn from its environment and on a “reward and punishment” based functions. YOU CAN READ MORE ABOUT REINFORCEMENT LEARNING HERE

A good robotic ecosystem needs geeks to both build robots and make them useful. We need some beautiful engineers to make our robots behave.

Our generation is the perfectly timed one to see the world of Robotics break that barrier and penetrate the lives of human just like computers did in the last century. It is high time that you should start building up your AI skills if you really want to be part of the journey and to be at the center of the revolution. You can have a basic idea of how the learning takes place by visualizing the image below.

Below I provide links of 2 videos for people who are interested in reinforcement learning in robotics .

The first video is an introduction and the second video is for people who have understood the Bellman Equation , Q functions and other reinforcement fundamentals