Efficient Coding in Motor Planning
The nervous system utilizes prior information to enhance the accuracy of perception and action. Prevailing models of motor control emphasize Bayesian models, which suggest that the system adjusts the current motor plan by integrating information from previous observations. While Bayesian integration has been extensively examined, those studies usually applied a highly stable and predictable environment. In contrast, in many real-life situations, motor goals change rapidly over time in a relatively unpredictable way, leaving it unclear whether Bayesian integration is useful in those natural environments. An alternative model that leverages prior information to improve performance is efficient coding, which suggests that the motor system maximizes the accuracy by dynamically tuning the allocation of the encoding resources based on environmental statistics. To investigate whether this adaptive mechanism operates in motor planning, we employed center-out reaching tasks with motor goals changing in a relatively unpredictable way, where Bayesian and efficient coding models predict opposite sequential effects. Consistent with the efficient coding model, we found that current movements were biased in the opposite direction of previous movements. These repulsive biases were amplified by intrinsic motor variability. Moreover, movement variability decreased when successive reaches were similar to each other. Together, these effects support the presence of efficient coding in motor planning, a novel mechanism with which the motor system maintains flexibility and high accuracy in dynamic environments.