MOSAIC Model for Sensorimotor Learning and Control

Abstract

Humans demonstrate a remarkable ability to generate accurate and appropriate motor behavior under many different and often uncertain environmental conditions. We previously proposed a new modular architecture, the modular selection and identification for control (MOSAIC) model, for motor learning and control based on multiple pairs of forward (predictor) and inverse (controller) models. The architecture simultaneously learns the multiple inverse models necessary for control as well as how to select the set of inverse models appropriate for a given environment. It combines both feedforward and feedback sensorimotor information so that the controllers can be selected both prior to movement and subsequently during movement. This article extends and evaluates the MOSAIC architecture in the following respects. The learning in the architecture was implemented by both the original gradient-descent method and the expectation-maximization (EM) algorithm. Unlike gradient descent, the newly derived EM algorithm is robust to the initial starting conditions and learning parameters. Second, simulations of an object manipulation task prove that the architecture can learn to manipulate multiple objects and switch between them appropriately. Moreover, after learning, the model shows generalization to novel objects whose dynamics lie within the polyhedra of already learned dynamics. Finally, when each of the dynamics is associated with a particular object shape, the model is able to select the appropriate controller before movement execution. When presented with a novel shape-dynamic pairing, inappropriate activation of modules is observed followed by on-line correction.

Imaging a cognitive model of apraxia: The neural substrate of gesture‐specific cognitive processes

“…Cerebellum plays also a significant role in the early phases of acquisition and planning of motor sequences [Doyon et al, 2002], and is known to participate in a wide variety of cognitive and emotional processes [e.g., see Marien et al, 2001;Middleton and Strick, 1998;Rapoport et al, 2000;Salman, 2002]. Moreover, a modular organization of internal models of tool manipulation has been recently reported in the cerebellum using fMRI [Imamizu et al, 2003], extending the predictions of the MOSAIC computational model [Haruno et al, 2001;Wolpert and Kawato, 1998] from to the "motor" to the "cognitive" cerebellum. In the present study, we suggest that the vermian cerebellar activation associated with the innervatory pattern stage merely reflects its contribution to movement regulation [Deiber et al, 1996;Jueptner et al, 1996;Kitazawa, 2002] and integration of simple movements into more complex ones [Ramnani et al, 2001;Thach, 1998] during actual movement production.…”

“…Alternative interpretations can be found in the motor control, especially computational, literature [e.g., Doya, 2000;Flanagan and Wing, 1997;Haruno et al, 2001;Schall and Bichot, 1998;Wolpert and Ghahramani, 2000;Wolpert and Kawato, 1998], which shows how sensory and motor experiences may interact in movement representation, generation, and production. However, probing such models was beyond the scope of the present study centred on current cognitive models of apraxia.…”

Robotics and neuroscience: A rhythmic interaction

“…The most celebrated concepts include the so-called internal model principle (see e.g. Haruno, Wolpert, and Kawato (2001), Kawato (1999), Mehta and Schaal (2002), Miall, Weir, Wolpert, and Stein (1993), Wolpert and Miall (1996) and Wolpert, Miall, and Kawato (1998)) and the theory of optimal control (see e.g. Scott (2004), Todorov (2004) and Todorov and Jordan (2002)).…”

Modeling of autonomous problem solving process by dynamic construction of task models in multiple tasks environment

“…MOSAIC model [D.M.Wolpert and M.Kawato. 1998;M.Haruno and D.M.Wolpert 2001] is composed of multiple modules in which a forward model and an inverse model are forming a unit. This system decides output of each module and learns each module in proportion as a prediction error of the forward model.…”