Sensorimotor decision-making with moving objects

Abstract

Moving is essential for us to survive, and in countless occasions we move in response to visual information. However, this process is characterized as uncertain, given the variability present both at the sensory and motor stages. A crucial question, then, is how to deal with this uncertainty in order for our actions to lead to the best possible outcomes.

Statistical decision theory (SDT) is a normative framework that establishes how people should make decisions in the presence of uncertainty. This theory identifies the optimal action as that which maximizes the expected reward (outcome) of the situation. Movement planning can be reformulated in terms of SDT, so that the focus is placed on the decisional component. Some experimental work making use of this theoretical approach has concluded that humans are optimal movement planners, while other has identified situations where suboptimality arises. However, sensorimotor decision-making within SDT has commonly eluded scenarios of interaction with moving objects. At the same time, the work devoted to moving objects has not focused on the decisional aspect. The present thesis aims at bridging both fields, with each of our three studies trying to answer different questions.

Given the spatiotemporal nature of situations with moving objects, we can plan our actions by relying on both temporal and spatial cues provided by the object. In Study I we investigated whether exploiting more one type of these visual cues led to a better performance, as defined by the reward given after each action. In our task we presented a target, which could vary in speed and motion time, approaching a line. Participants responded to stop the target and were rewarded according to its proximity to the line. Responding after the target crossed the line was penalized. We discovered that those participants planning their responses based on time-based motion cues had a better performance than those monitoring the target’s changing spatial position. This was due to the former approach circumventing a limitation imposed by the resolution of the visual system. We also found that viewing the object for longer favored time-based responses, as mediated by longer integration time. Finally, we used existing SDT models to obtain a reference of optimality, but we defend that these models are limited to interpret our data.

Study II built on our previous findings to explore whether the use of temporal cues could be learnt. We took our previous paradigm and adapted it so that reward was manipulated after each task in order to foster exploiting temporal information. There was no evidence for learning taking place, since participants using temporal cues did so from the start of the experiment. Whether other methods reward can shape the use of certain cues, and why some people naturally tend to make more use of temporal information, still remain elusive.

Study III deepened our knowledge on which variability people consider when planning their responses. We hypothesized that the reason why people are suboptimal (as defined by SDT) in many situations is because they represent only their measurement variability, roughly equivalent to the execution noise, while excluding the variability created by sudden changes in their planning. We took previous data and used a Kalman filter to extract each participant’s measurement variability. We then used it to compute SDT-derived optimal responses, and discovered that they explained well our data, giving support to our hypothesis. We also found evidence for participants using the information provided by reward both to avoid being penalized and to choose the point at which to stabilize their responses.

Taken together, our experimental work presents interaction with moving objects as a complex set of situations where different information guides our response planning. Firstly, visual cues of different origin. Secondly, our variability, coming from many sources, some of which may not be considered. Finally, the outcomes related to each action.

Moure’s és essencial per a la nostra supervivència, i en incomptables ocasions ens movem en resposta a informació visual. Tanmateix, aquest procés és incert, donada la variabilitat present tant a l'estadi sensorial com en el motor. Una pregunta crucial, doncs, és com gestionar aquesta incertesa perquè les nostres accions portin a les millors conseqüències possibles.

La teoria de la decisió estadística (Statistical decision theory, SDT) és un marc teòric normatiu que estableix com la gent hauria de fer decisions en presència d'incertesa. Aquesta teoria identifica l'acció òptima amb aquella que maximitza la recompensa (entesa com a conseqüència) esperada de la situació. La planificació del moviment pot ser reformulada en termes de SDT, de tal manera que s’emfatitza el component decisional. Diferents treballs experimentals que han fet servir aquesta aproximació teòrica han conclòs que els humans som planificadors de moviment òptims, mentre que altres han identificat situacions on la suboptimalitat sorgeix. No obstant això, la presa de decisions sensoriomotora des de SDT normalment ha ignorat escenaris que requereixen d'interacció com objectes en moviment. Alhora, els treballs dedicats als objectes en moviment no s'han centrat en l'aspecte de decisió. La present tesi es proposa acostar els dos camps, amb cada un dels nostres tres estudis intentant respondre diferents preguntes.

L’Estudi I descobrí que, per planificar les nostres decisions, fer servir informació temporal portà a un millor rendiment que fer servir informació espacial, i això fou facilitat per veure l'objecte durant més temps. També vam criticar la limitació de certs models d’SDT per interpretar els nostres dades. L'Estudi II intentà promoure l'ús d'informació temporal, tot i que no s’aconseguí fomentar l’aprenentatge. Finalment, l’'Estudi III trobà que la raó per la qual la gent és subòptima en moltes situacions es deu al fet que representa només la seva variabilitat de mesura, més o menys equivalent al soroll d'execució, mentre que s'exclou la variabilitat creada per sobtats canvis en la planificació de la resposta. També trobàrem que els participants van usar la informació donada per la recompensa tant per evitar ser penalitzats com per escollir el punt on estabilitzar les seves respostes.