This repository contains a MATLAB simulation of a planar dual-arm robot featuring antagonistic joint actuation and equipped with proprioceptive, tactile, and visual sensors. The data generated by the robot is used to discover a set of emergent behavioral primitives through the analysis of dynamic functional connectivity among the sensory signals.
Citation: Diaz Ledezma, F., Marcel, V., & Hoffmann, M. (2025). Unsupervised Discovery of Behavioral Primitives from Sensorimotor Dynamic Functional Connectivity. ArXiv, abs/2506.22473.pdf, youtube video.
Everything works on MATLAB R2023b. Requirement:
- Robotics System Toolbox for the robot dynamics model used in Simulink
- Peter Corkes Matlab Robotics Toolbox RTB10.4.mltbx
- Other external packages are already contained in the external toolbox directory
The Live Script visuosomatosensory_exploration_ICDL.mlx is used to instantiate the agent and run the simulation. It also
- Computes the instantaneous mutual information matrix
- Runs the infinite relational model
- Performs non-negative matrix factorization It may take significant time to run each of the three procedures.
To replicate the exact results in the paper, download this MAT file, load it into MATLAB's workspace, and execute the cells in the Live Script results_ICDL.mlx
Send message to Fernando Diaz Ledezma: [email protected]
- Extension to a simulated iCub robot in Python is underway.
If you wish to collaborate, please contact Matej Hoffmann at [email protected]
- Fernando Diaz Ledezma
- Valentin Marcel
- Matej Hoffmann
CC-BY-4.0