A sensorimotor paradigm for Bayesian model selection.

Area of Science:

• Neuroscience
• Motor Control
• Cognitive Science

Background:

• Sensorimotor control relies on predictive internal models for navigating uncertainty.
• Humans learn task-specific models and extract common structures between tasks.
• Selecting between models and parameters is crucial for motor adaptation.

Purpose of the Study:

• Investigate how the motor system selects between different internal models and parameters.
• Test Bayesian statistics as a model for sensorimotor selection behavior.
• Develop an experimental paradigm to disentangle model and parameter variables.

Main Methods:

• Designed a 3D virtual reality sensorimotor task with visuomotor shifts.
• Mapped task dimensions to model and parameter variables.
• Utilized neutral probe trials to directly assess model selection.

Main Results:

• Bayesian statistical model selection better explained subject behavior than non-probabilistic heuristics.
• Demonstrated the ability to independently probe model and parameter variables.
• Provided evidence for sophisticated model selection in sensorimotor adaptation.

Conclusions:

• Bayesian inference plays a significant role in sensorimotor model selection.
• The experimental design effectively separates model and parameter influences.
• This framework facilitates further research into sensorimotor learning and adaptation.