Abstract
How does the evidence for the trajectory of a point moving in noise evolve over time? Observers were asked to track a linear motion trajectory presented in Brownian noise either by indicating the direction with one of 8 keys on a numeric keypad or by manually tracking the trajectory with a stylus on a digitizing tablet. The keypad task required a response after stimulus offset and was repeated at different stimulus durations to determine the duration that leads to reliable performance. The manual-tracking task had a fixed duration of 1000 ms and observers were asked to actively track the trajectory during stimulus presentation.
Data for three observers show that evidence for the trajectory increases with time, and that the time required for criterion performance is longer for high-noise than no-noise conditions. Interestingly, the amount of extra time needed to maintain performance in high-noise conditions was the same for both the keypad and manual tracking experiments, 60 ms. This additional time was constant despite an order of magnitude difference in the total stimulus duration before a response: 30 ms for the keypad and 300 ms (which included processing and decision delays as well as motor preparation time) for the manual tracking experiments.
More importantly, in both tasks the direction errors in the noise task form a broad distribution at short stimulus durations/latencies (indicating uncertainty about the trajectory), and the distribution narrows quickly over time to a steady state. The steady-state distribution peaks sharply at the true trajectory direction, consistent with a predictive filter model that incorporates local motion responses up to that point in time and feedback that favors smoothness.
Supported by NSF grant BCS0347051 to PV and NIH grants R21 EY017003