Event-driven figure-ground organisation model for the humanoid robot iCub.

Area of Science:

• Robotics
• Computer Vision
• Neuroscience

Background:

• Figure-ground organization is crucial for environmental interaction but computationally intensive with current methods.
• Classical and deep learning approaches to segmentation demand significant resources, particularly for training.

Purpose of the Study:

• To develop a bio-inspired, biologically plausible perception system for the neuromorphic robot iCub.
• To leverage event-driven vision for efficient figure-ground segmentation.

Main Methods:

• A hierarchical, biologically inspired architecture was implemented.
• Event-driven cameras were utilized to minimize data redundancy and computation.
• The system was assessed qualitatively and quantitatively in simulations and on the iCub robot.

Main Results:

• The system successfully segmented objects in diverse real-world scenarios.
• Performance was comparable to frame-based methods on simple stimuli and benchmark datasets.
• Event-driven vision demonstrated reduced data redundancy and computational load.

Conclusions:

• The bio-inspired, event-driven system offers an efficient alternative for figure-ground segmentation.
• This model enhances hybrid systems by processing relevant data in Regions of Interest (ROI).
• Enables low-latency autonomous robotic applications.