Abstract
The Online Processing of Dynamics Model (OPoD) offers a novel account of how the visual system estimates mass in dynamic scenes by treating mass and velocity as jointly constrained variables. Rather than relying on post hoc inference, OPoD characterizes mass estimation as a process in which cues to mass-such as apparent volume- are combined with sensed velocities to yield momentum-based expectations about post-collision motion. These momentum-based expectations of mass bias the estimation of post-collision velocity and those biased velocities feedback to reshape mass judgments, revealing that mass and velocity are jointly encoded in a shared momentum-like representation where each constrains the other. Across three experiments, we test two core predictions of our OPoD model: (1) that initial impressions of object mass bias perceived post-collision velocity, and (2) that biases in velocity in turn bias relative mass judgments. Each prediction is confirmed empirically. We formally compare OPoD against an adapted version of the noisy Newton model, in which mass priors are scaled to visible volume. Even with this extension and task-specific parameter tuning, the noisy Newton model does not provide a unified account of the mass and velocity biases evident in the data. OPoD consistently achieves better quantitative fits (higher log-likelihood, lower RMSE, stronger correlations) and provides a mechanistic, ecologically grounded complement to probabilistic simulation frameworks-offering a perceptual mechanism for how physical expectations are integrated and co-constrained during an event.