A pilot study by Subramanian et al. on Markov decision problem task decomposition by humans revealed that participants break down tasks into both short-term subgoals with a defined end-condition (such as "go to food") and long-term considerations and invariants with no end-condition (such as "avoid predators"). In the context of Markov decision problems, behaviors having clear start and end conditions are well-modeled by an abstraction known as options, but no abstraction exists in the literature for continuous constraints imposed on the agent's behavior. We propose two representations to fill this gap: the state constraint (a set or predicate identifying states that the agent should avoid) and the state-action constraint (identifying state-action pairs that should not be taken). State-action constraints can be directly utilized by an agent, which must choose an action in each state, while state constraints require an approximation of the MDP’s state transition function to be used; however, it is important to support both representations, as certain constraints may be more easily expressed in terms of one as compared to the other, and users may conceive of rules in either form. Using domains inspired by classic video games, this dissertation demonstrates the thesis that explicitly modeling this negative policy information improves reinforcement learning performance by decreasing the amount of training needed to achieve a given level of performance. In particular, we will show that even the use of negative policy information captured from individuals with no background in artificial intelligence yields improved performance. We also demonstrate that the use of options and constraints together form a powerful combination: an option and constraint can be taken together to construct a constrained option, which terminates in any situation where the original option would violate a constraint. In this way, a naive option defined to perform well in a best-case scenario may still accelerate learning in domains where the best-case scenario is not guaranteed.

La perception visuelle de l’espace est largement déterminée par les capacités visuelles des individus. Cependant, la recherche sur l’influence de variables dites non visuelles semble indiquer une importance déterminante des dimensions corporelle et émotionnelle sur la perception visuelle. Dans cette thèse, nous nous sommes intéressés à la perception des distances et avons tenté de comprendre en quoi nos perceptions pouvaient refléter la nature de nos interactions sensorimotrices avec notre environnement. Dans une première étude nous avons montré que des contraintes d’actions, tel que le coût énergétique, générées par les dispositions de l’environnement, pouvaient avoir une influence sur la perception de distances allocentriques. Dans une seconde étude, nous avons tenté de mettre en évidence le rôle du processus d’intégration sensorimotrice dans des tâches d’évaluation de distances, ainsi que de comprendre l’influence des modalités non visuelles, telles que l’audition et la motricité, dans ce même type de tâche. Enfin, dans une troisième étude, nous nous sommes intéressés au concept de fluence – i.e., information métacognitive renseignant le système sur la qualité de ses interactions dans son environnement – et proposons l’idée selon laquelle ce signal contenu dans le flux perceptif permettrait d’expliquer différentes variations perceptives liées aux contraintes de nos actions. Finalement, cette thèse défend une approche unifiée de la perception, selon laquelle la limite entre les concepts d’action et de perception devrait être repensée afin de rendre compte de la nature sensorimotrice de nos connaissances. / Visual perception of space is mainly known as depending upon one’s vision capacity. However, research about the influence of non-visual variables seems to indicate that the body also plays an important role in visual perception. In this Ph.D. thesis, we support the idea that distance perception has to be studied as a function of the sensorimotor interactions between the individuals and their environment. In the first study, we showed that action constraints in the environment such as energetic cost could influence allocentric distance perception. In the second study, we focused on the role of the sensorimotor integration process in distance perception tasks, as well as the influence of non-visual variables such as audition and motor activity. In the third study, we investigated the phenomenology of perception, and more precisely in the relative fluency of motor activity. In other words, we focused here on the metacognitive feedback that emerges from the quality of the interactions with the environment. We highlighted that such a metacognitive signal could explain the influence of action constraints on distance perception. Finally, this work provides strong supports to the idea of an integrative approach of perception according to which the theoretical boundary between perception and action is questioned by the sensorimotor nature of our knowledge.