A Neural Model for Exploration and Learning of Embodied Movement Patterns

Abstract

Despite increased interest in the study of human motor development among researchers ranging brain scientists to roboticists, many aspects of the mechanisms involved remain to be clarified. Talking a synthetic approach to development, to extract the property of the mechanism for a new robot controller, we propose two movement learning properties in the human being: (1) compression of redundant motor commands and (2) mapping from sensors to motors in the coupling of the controller, the body, and the environment. To prove the feasibility of our proposition, we constructed a neural model having essential biological features. In a series of experiments with a simple body model, rhythmic movement (to be learned in early infancy) is explored and correctly learned; moreover entrainment is observed. Our results suggest that our model can learn rhythmic movement, confirming a first step towards understanding of human development.

Chaotic Exploration and Learning of Locomotion Behaviors

“…Recent robotics studies have demonstrated that chaotic neural networks can indeed power the self-exploration of brain-body-environment dynamics in an embodied system, discovering stable patterns that can be incorporated into motor behaviors (Kuniyoshi & Suzuki, 2004;Kinjo, Nabeshima, Sangawa, & Kuniyoshi, 2008;Pitti et al, 2010).…”

“…The biomechanical system was modeled as a series of redundant muscles acting on a joint, and information on the muscle combinations for any discovered coherent motor patterns was engraved on the model cortices as a sensorimotor representation. Later work (Kinjo et al, 2008) demonstrated the learning and replay of a motor pattern by adding a simple perceptron with a backpropagation learning on top of the previously learned sensorimotor maps. They showed that the representative power of the self-organized sensorimotor maps can greatly simplify the nontrivial sensorimotor learning problem into a simple mapping between the sensor and motor maps, but the learning pattern was manually fed to the system during learning; hence, it cannot be regarded as an example of an autonomous and goal-directed exploration-learning scheme.…”

General Principles of Neurorobotic Models Employing Entrainment and Chaos Control

“…In addition to entrainment of brain, body, and environment, internal components of the brain (modeled as neural oscillators in the studies we review here) must also be mutually entrained or coupled to one another in order for the system as a whole to enact complex dynamical behavior. We see a key example of this coupled chaotic field models (Kuniyoshi and Suzuki, 2004; Pitti et al, 2005, 2010; Kuniyoshi and Sangawa, 2006; Kinjo et al, 2008; Mori and Kuniyoshi, 2010), which connect chaotic elements together into a complex system of interacting components.…”

“…In Shim and Husbands (2012), discovered patterns are memorized by wiring initially disconnected oscillators using ‘adaptive synchronization.” In Sussillo and Abbott (2009), so-called “FORCE learning” is used to modify synaptic strengths and stabilize chaotic spontaneous activity to desired activity patterns. In Kinjo et al (2008), two aspects of learning are modeled: the compression of redundant motor commands and the mappings which couple controller, body, and environment. In Kuniyoshi and Sangawa (2006) models of sensory and motor cortex based on a dynamic variant of Kohonen self-organizing maps (Goodall et al, 1997; Chen, 1998) are employed, and associations are learned between sensory and motor areas and via cortical areas and “spinal” CPGs via Hebbian learning.…”

Darwinian Dynamics of Embodied Chaotic Exploration

“…Kuniyoshi and Sangawa [7] made the important suggestion that chaotic dynamics underpin crucial periods in animal development when brain-body-environment dynamics are explored in a spontaneous way as part of the process of acquiring motor skills. A few robotics studies have demonstrated that chaotic neural networks can indeed power the self-exploration of brain-body-environment dynamics in an embodied system, discovering self organized patterns that can be incorporated into motor behaviors [8,6,12].…”