Is it me or the train moving? Humans resolve sensory conflicts with a nonlinear feedback mechanism in balance control.

Abstract

Humans use multiple sensory systems to estimate body orientation in space. Sensory contributions change depending on context. A predominant concept for the underlying multisensory integration (MSI) is the linear summation of weighted inputs from individual sensory systems. Changes of sensory contributions are typically attributed to some mechanism explicitly adjusting weighting factors. We provide evidence for a conceptually different mechanism that performs a multisensory correction if the reference of a sensory input moves in space without the need to explicitly change sensory weights. The correction is based on a reconstruction of the sensory reference frame motion (RFM) and automatically corrects erroneous inputs, e.g., when looking at a moving train. The proposed RFM estimator contains a nonlinear dead-zone that blocks corrections at slow velocities. We first demonstrate that this mechanism accounts for the apparent changes in sensory contributions. Secondly, using a balance control model, we show predictions of specific distortions in body sway responses to perturbations caused by this nonlinearity. Experiments measuring sway responses of 24 subjects (13 female, 11 male) to visual scene movements confirmed these predictions. The findings indicate that the central nervous system resolves sensory conflicts by an internal reconstruction of the cause of the conflict. Thus, the mechanism links the concept of causal inference to shifts in sensory contributions, providing a cohesive picture of MSI for the estimation of body orientation in space.Significance statement How the central nervous system (CNS) constructs body orientation in space from multiple sensory inputs is a fundamental question in neuroscience. It is a prerequisite to maintain balance, navigate and interact with the world. To estimate body orientation, the CNS dynamically changes the contribution of individual sensory inputs depending on context and reliability of the cues. However, it is not clear how the CNS achieves these dynamic changes. The findings in our study resolve major aspects of this question. Importantly, the proposed solution using nonlinear multisensory feedback contrasts with traditional approaches assuming context-dependent gain-scaling of individual inputs. Thus, our findings demonstrate how complex, intelligent, and unintuitive behavior can emerge from a comparably simple nonlinear feedback mechanism.