The crawler robot is a simple and low-cost physical platform for the initial study of reinforcement learning algorithms. It consists of a 2 DoF planar arm actuated by servomotors mounted on a mobile base with 2 free wheels .
The reinforcement learning algorithm used for the robot to find a sequence of actions that provide locomotion by its own is known as Q-learning. The agent starts by executing random movements with his arm and when some sequence manages to mobilize the platform, the robot receives a positive reward. It is the repetition of this reward process that allows the robot to strengthen the movements that give it consistent locomotion and ignore the ones that do not.
- Ardunio Uno board
- Matlab 2018
- 2 hobby servomotors
- Optical enconder
- Bluetooth module
- The robotic agent performing random movements in order to learn an optimal policy through negative feedback.
- The robotic agent executing a locomotion strategy through a set of self-learned coherent movements