A cerebellar model for predictive motor control tested in a brain-based device

Abstract

The cerebellum is known to be critical for accurate adaptive control and motor learning. We propose here a mechanism by which the cerebellum may replace reflex control with predictive control. This mechanism is embedded in a learning rule (the delayed eligibility trace rule) in which synapses onto a Purkinje cell or onto a cell in the deep cerebellar nuclei become eligible for plasticity only after a fixed delay from the onset of suprathreshold presynaptic activity. To investigate the proposal that the cerebellum is a generalpurpose predictive controller guided by a delayed eligibility trace rule, a computer model based on the anatomy and dynamics of the cerebellum was constructed. It contained components simulating cerebellar cortex and deep cerebellar nuclei, and it received input from a middle temporal visual area and the inferior olive. The model was incorporated in a real-world brain-based device (BBD) built on a Segway robotic platform that learned to traverse curved paths. The BBD learned which visual motion cues predicted impending collisions and used this experience to avoid path boundaries. During learning, the BBD adapted its velocity and turning rate to successfully traverse various curved paths. By examining neuronal activity and synaptic changes during this behavior, we found that the cerebellar circuit selectively responded to motion cues in specific receptive fields of simulated middle temporal visual areas. The system described here prompts several hypotheses about the relationship between perception and motor control and may be useful in the development of general-purpose motor learning systems for machines. cerebellum ͉ computational model ͉ eligibility trace ͉ middle temporal visual area A recent theory of the cerebellum (1), consistent with much of the neurophysiological, behavioral, and imaging data regarding motor control, proposes that the cerebellum learns to replace reflexes with a predictive controller. This predictive controller produces a correct motor control signal earlier than less adaptive reflex responses. Numerous adaptive cerebellar functions are subject to this type of control, including eye-blink conditioning, the vestibular-ocular reflex, smooth pursuit eye movement, spinal nociceptive withdrawal reflex, grip force adjustments, arm movements, and saccadic eye movements. At present, however, the mechanisms responsible for this predictive capability are still debated. Proposals for such mechanisms include delay lines, spectral timing, oscillators, or dynamic recurrent activity in granule cells (for a review, see ref.2).It has been suggested that reflexive motor commands are used as error signals for the predictive controller and that they are delivered to the cerebellum by way of climbing fibers from the inferior olive (IO) (1, 3). Synaptic eligibility traces in the cerebellum have recently been proposed as a specific mechanism for such motor learning (4). Based on motor timing studies, Medina et al. (4) proposed an eligibility trace that is triggered by motion onset and tha...

Chained learning architectures in a simple closed-loop behavioural context

“…This differs from our approach because it is supervised learning and the learning also does not take place in a complete closed-loop setting since the output of the learner is not used to drive the car. This differs from a recent study by McKinstry et al (2006), who were able to close the loop and derive path-following behaviour in a robot driven by a complex multilayer neuronal system supposed to mimic parts of the cerebellar system. The system learns, as in our case, reflex avoidance.…”

Communicating emotions and mental states to robots in a real time parallel framework using Laban movement analysis

“…Brain inspired robots have been used for investigating animal locomotion and motor control, [11][12][13][14]9], to learn to avoid obstacles [15,16], produce accurate vision functions [17][18][19] generate adaptive arm movements [17,20,13,9], perform (ratlike) learning and memory tasks [21,14,[22][23][24], or to emulate the human or rodent reward and value systems [25,26].…”

Modelling of syntactical processing in the cortex

“…McKinstry et al, 2006). Our cortical language system can better be understood in the context of a larger model that covers many cortical areas and integrates language understanding, visual object recognition, visual attention, and action planning.…”