Let’s talk today about reinforcement learning!
And to illustrate this theoretical topic, we’ve attached a video!
If you need a short introduction about what this technical term means: It is an established approach in the field of machine learning, which is inspired by natural learning processes. An agent autonomously learns to follow a specific “policy,” – meaning a strategy triggered by rewarding/ not rewarding actions of the environment.
At ARTI we experiment with the more advanced method of adversarial reinforcement learning, the future gold standard for machine learning, to test our software approaches (see more in the video). With this approach, we train our software on policies regarding the capability to follow objects and paths autonomously and without collision.
Reinforcement learning to us has several advantages when it comes to the improvement of our AI software kits:
- Automatic parameter tuning
- Detection of absolute limitations
- Behaviors beyond human performance
In the first half of the video below, you can observe how the robot learns to reach a target point. In order to speed up the learning process, four robots perform this simultaneously. In the second half, the mission is to follow a moving target and avoid collision with obstacles. The blue robot chases the green one, trying to escape the blue one. This learning process is still ongoing and will show improvement in the future.