The Cerebellum as a Predictive Model of the Motor System: A Smith Predictor Hypothesis

Summary

The performance of motor systems with large feedback delays can be significantly enhanced by the use of internal predictive representations of the motor apparatus. The cerebellum is a likely site for these internal models, and we show that ataxic patients appear to have reduced awareness of the hand position during movement, suggesting that a sensory predictor within the cerebellum is impaired. We recently suggested that the cerebellum holds two types of neural model which together form a ‘Smith Predictor’. One is a model of the motor apparatus (limbs and muscles) which provides a rapid prediction of the sensory consequences of each movement. The other model is of the time delays in the feedback control loop (conductance delays, muscle latencies, sensory processing). This delays a copy of the rapid prediction, so that it can be compared with actual sensory feedback; any errors are then used both to correct the movement and to update the internal representations of the motor apparatus. We propose mechanisms by which both parts of the Smith Predictor could be formed within the cerebellum, and present a neural network simulation based on these ideas.