Jordi Grau Moya

Alumni of the Research Group Sensorimotor Learning and Decision-Making

Main Focus

In previous studies our lab has investigated the effects of risk-sensitivity and two-player interaction on motor control. Both factors are important neuroeconomic constraints both experimentally and in normative models of acting. However, important open questions remain about how uncertainty over different models (‘ambiguity’ or ‘model uncertainty’) is integrated by human subjects, how sensory uncertainty is affected by risk-attitudes and how uncertainty over different states of information in two-player interactions (types of players, cost functions, structure of the game, etc.) affects sensorimotor behaviour.

1: What is the effect of risk-sensitivity on sensory integration?

In this series of experiments, we will investigate the impact of risk-sensitivity on Bayesian integration in motor control. In the previous studies it was examined how risk-sensitivity influences the selection of actions and control policies in the face of uncertainty in motor control tasks – it was not investigated however how risk-sensitivity impacts on estimation. In a risk-neutral controller, estimation and control processes are decoupled by the certainty-equivalence principle. In contrast, in a risk-sensitive controller the processes become coupled, so that the cost function can bias estimates of sensory input. In our experiment, subjects will perform reaching movements that are associated either with an explicitly given cost or with an implicit cost given by a resistive force. The level of acuity of the visual feedback of their movements will be varied. We can then examine whether their coupling of estimation and control are biased in accordance with the predictions of a risk-sensitive control scheme (for example, by first fitting the risk-parameter as in and then predicting the estimation bias).

2: What is the effect of model uncertainty (ambiguity) on sensorimotor learning and control?

In this series of experiments, we will investigate the potential difference between risk (known uncertainty) and ambiguity (unknown uncertainty, model uncertainty) in motor control. First, we will translate the famous Ellsberg paradox into a motor control task similar to our first task. The importance of the Ellsberg paradox is that it violates expected utility theory, because most subjects strictly prefer environments with known uncertainty (e.g. an urn with 50 red and 50 blue balls) as opposed to unknown uncertainty (e.g. an urn with unknown composition). In our experiment, subjects will be instructed to move towards a target and try to hit it. We will fit the size of the target such that subjects have 50% chance to hit the target. Additionally we will introduce ambiguous targets where the actual size of the target can only be partially perceived. After they are trained we can then give subjects a choice between the risky target with known hitting probability and an ambiguous target with unknown size. We will vary the amount of information that we reveal about the ambiguous target to investigate how ambiguity influences their choice and whether this choice behaviour is similar to choices made in the cognitive version of the Ellsberg paradox.

3: What is the effect of uncertain information on sensorimotor learning in two-player motor interactions?

In this series of experiments, we will extend our previous game-theoretic studies in motor control by investigating the impact of incomplete information, i.e. missing information about the “type” of the other player. In our previous studies this problem was not addressed. Here we will investigate how changing the degree of information has an impact on the strategies chosen by the players. Modelling these kinds of games requires the application of Bayesian game theory. We will use the experimental setup described in where subjects will be making reaching movements and experience resistive forces depending on their coupled cost functions. We will manipulate the following variables. At the beginning of trials we will display information about the cost function and / or the other players cost function. If the trial is repeated players can learn empirical frequencies about the probabilities with which their cost function type and the type of the other player are assigned. We will then compare their action choices to the ones predicted by Bayesian game theory. In particular, the risk-attitude of an agent also captures how this agent is pessimistic or optimistic with respect to its environment. We will test the effect of risk-attitude in cooperative games where optimistic attitudes of players can lead to higher levels of cooperation. To test this, subjects will play in pairs in games such as stag hunt. We will introduce computer players to manipulate the risk-attitude of the participants. Then we will test how risk-attitude affects cooperation.

Curriculum Vitae


  • from 2011 PhD student, Max Planck Institute for Biological Cybernetics & Intelligent Systems.
  • 2011. Industrial Engineer, Universitat Politecnica de Catalunya, Terrassa, Spain.

Research Experience

  • from 2011 PhD student, Max Planck Institute for Biological Cybernetics & Intelligent Systems.
  • 2011           Final project, DONLL Research group, Terrassa, Spain.
  • 2010 - 2011 MCIA Research Center (UPC) , Terrassa, Spain.
  • 2009 - 2010 Reserch assistant, LABSON Laboratory (UPC) Terrassa, Spain.
Go to Editor View