Sustained sensorimotor control as intermittent decisions about prediction errors: computational framework and application to ground vehicle steering

Abstract

A conceptual and computational framework is proposed for modelling of human sensorimotor control and is exemplified for the sensorimotor task of steering a car. The framework emphasises control intermittency and extends on existing models by suggesting that the nervous system implements intermittent control using a combination of (1) motor primitives, (2) prediction of sensory outcomes of motor actions, and (3) evidence accumulation of prediction errors. It is shown that approximate but useful sensory predictions in the intermittent control context can be constructed without detailed forward models, as a superposition of simple prediction primitives, resembling neurobiologically observed corollary discharges. The proposed mathematical framework allows straightforward extension to intermittent behaviour from existing one-dimensional continuous models in the linear control and ecological psychology traditions. Empirical data from a driving simulator are used in model-fitting analyses to test some of the framework’s main theoretical predictions: it is shown that human steering control, in routine lane-keeping and in a demanding near-limit task, is better described as a sequence of discrete stepwise control adjustments, than as continuous control. Results on the possible roles of sensory prediction in control adjustment amplitudes, and of evidence accumulation mechanisms in control onset timing, show trends that match the theoretical predictions; these warrant further investigation. The results for the accumulation-based model align with other recent literature, in a possibly converging case against the type of threshold mechanisms that are often assumed in existing models of intermittent control.

A quantitative driver model of pre-crash brake onset and control

“…The computational framework adopted here (Markkula, 2014;Markkula et al 2017) posits that:  Driving control can be regarded as a series of intermittent, open loop control adjustments (motor primitives).  The timing of new adjustments is determined by a process of evidence accumulation.…”

“…Markkula et al (2016) suggested that their findings from SHRP 2 naturalistic driving data could be explained if the driver's brake initiation is not solely related to the crossing of a looming threshold, but to noisy evidence accumulation of looming and other perceptual input over time, a type of mechanism for which there is much support from laboratory tasks in psychology and neuroscience (Gold & Shadlen, 2007). Combining this model mechanism with other well proven neuroscientific concepts, especially motor primitives (Giszter, 2015) and prediction of sensory outcomes of motor actions (Crape & Sommer, 2008), Markkula and colleagues (Markkula, 2014;Markkula, Boer Romano & Merat, 2017) have developed a computational framework for driver control behaviour. This paper describes, for the first time, an application of this framework to braking behaviour, more specifically braking in critical scenarios.…”

At the Zebra Crossing: Modelling Complex Decision Processes with Variable-Drift Diffusion Models

“…Yet many sensorimotor decisions take place in the context of a continuous stream of varying sensory information. Models with variable drift rate, which we will refer to here as variable-drift diffusion models (VDDMs), have been successful in the vehicle driving context, accounting well for driver brake responses to the time varying visual looming of an approaching vehicle (Xue, Markkula, Yan, & Merat, 2018) as well as for steering responses during lanekeeping (Markkula, Boer, Romano, & Merat, 2018).…”

Simulating the effect of cognitive load on braking responses in lead vehicle braking scenarios

“…The present model is based on the evidence accumulation framework developed by Markkula [40,41] and also incorporated key principles from the GAT model mentioned above [33][34][35][36]. Similar computational implementations of the GAT model have previously been developed for laboratory tasks such as the Stroop task [34,35].…”