Bayesian inference in arm posture perception.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Human Motor Control

Background:

• The brain computes limb position using noisy sensory information from mechanoreceptors.
• Proprioception, the sense of limb position, is believed to involve probabilistic processing and Bayesian inference.

Purpose of the Study:

• To test the hypothesis that arm posture perception follows Bayes' rule by integrating noisy sensory data with prior beliefs.
• To compare a Bayesian model with a Gaussian prior against a model with a uniform prior for explaining arm posture perception errors.

Main Methods:

• Developed a computational model of arm posture perception as a probabilistic kinematic chain with Gaussian noise and a Gaussian prior.
• Conducted a virtual reality (VR)-based posture matching task where participants aligned their unseen arm to a target posture.
• Used maximum likelihood estimation to fit the Bayesian model to behavioral data and estimate prior parameters.

Main Results:

• Individual participants exhibited idiosyncratic biases in matching their unseen arm to target postures.
• The Bayesian model incorporating a Gaussian prior significantly outperformed a uniform prior model in explaining response biases and variance.
• Participant-specific priors were estimated, reflecting individual differences in arm posture perception.

Conclusions:

• Arm posture perception is well-modeled as a Bayesian computation integrating sensory input with non-uniform, participant-specific priors.
• This Bayesian framework provides new insights into the nature of proprioceptive computations and clarifies biases in spatial awareness.
• The findings highlight the role of prior beliefs in shaping our sense of limb position.