Teaching a humanoid robot to navigate using Artificial Intelligence

In this blog, we will see how a humanoid robot learns to navigate to its goal using Artificial intelligence (AI) in a simulation. The AI that we will be using to teach the Humanoid is called Reinforcement learning.

What is Reinforcement Learning?

Reinforcement learning is a form of machine learning which is primarily concerned with how an agent can learn to act optimally in an unknown environment. The agent, called the learner (Humanoid in our case), interacts with its environment through actions. The learner’s goal is to find a policy for selecting actions that maximize its total reward over time.

The humanoid starts with a set of possible policies and evaluates them by simulating the consequences of each policy and comparing them against the current best policy, called the current model-predictor. The learner selects the action from among those predicted by the current model predictor that leads to a higher total reward than any other predicted action would have led to.

Let's train the Humanoid to navigate

The humanoid learns to find its goal using reinforcement learning in a Unity3D simulation. The environment is being created using mlagents, where the goal of the humanoid is to…