hesamazj/MusicianHand: Musician Hand's Repository

Project Overview The "Musician Hand" project represents an end-to-end learning system designed to imbue a robot with a musical skill. Instead of explicitly programming the robot's movements for each note, we leverage principles of self-supervised learning: Exploratory Babbling: The robot performs random, diverse actions (limb activations) for a short period (2 minutes). Perception-Action Data Collection: During babbling, the system simultaneously records the robot's commanded actions and its sensory output (e.g., the sound produced by its movements). Inverse Model Learning: A neural network is trained to learn the inverse dynamics – mapping desired musical outputs (e.g., specific note characteristics from a melody's spectrogram) back to the required robot limb activations that produced those sounds during babbling. Melody Playback (Deployment): Once the inverse model is trained, a target 4-note melody is fed into the model, which then predicts the necessary robot activations to reproduce the melody. These activations are sent to the robot in real-time. This approach simulates how biological systems learn motor skills through exploration and sensory feedback, enabling the robot to "understand" how its actions translate into musical outcomes.