To develop, and implement in robots, a biologically-inspired mathematical model of how spatiotemporal patterns of tactile information can be efficiently encoded and recalled for the selection of action and control of behavior. The core of the proposed method is a computational framework, that mimics the compression, chunking and consolidation and recall functions of mammalian episodic memory. The approach is based on a family of Bayesian, latent variable models referred to as “Deep Gaussian Process” (DGP) that offers a probabilistic framework for modelling complex data in (un/semi)-supervised mode, and is therefore well-suited as a platform for perception and learning in robotics. This framework has already been validated for three aspects of robot perception: face recognition, arm/hand action recognition and passive touch gesture recognition using the iCub humanoid robot (FP7 WYSIWYD project). Results show the capacity of our proposed method for integration of sensory data, adaptability and accurate perception in real-time. The current project will extend this approach to active touch for the control of tactile object exploration and manipulation. The principled handling of uncertainty in the perception and prediction stages, through the Bayesian probabilistic formulation, constitutes a promising path to the development of applied systems for assistive and industrial robotics that will be explored during the secondment part of the studentship.

A probabilistic framework for integration, storage and recall of tactile memories using generative Gaussian process models inspired by perceptual coding, learning and episodic memory systems in the mammalian brain. The system will be evaluated in iCub robot as a contributor to multi-sensory guidance of active sensing, action selection, and object exploration/manipulation. Results will also be generalized from the iCub to more conventional industrial or assistive robotic systems.