Create your Ernest

Learn to create your own self-programming agent and robot in my course on developmental learning. French original version available here.

Lyon Startup Weekend

We won the jury's favorite prize ("coup de coeur du jury", 4^th prize) at the Lyon Startup Weekend ! The IDEAL team from left to right: Renaud Detaille, Raphaël Cazorla, Olivier Georgeon, Séverin Bruhat, and Julien Millet. Great thanks to the organizing team for this crazy unforgettable weekend. We love you too !

Watch the  finale video (our presentation from time 2:47:37 to 2:56:20, low audio quality French).

Inferring actions and observations from interactions

Garnier J., Georgeon O., and Cordier A., 2013. Inferring Actions and Observations from Interactions. In the Second Workshop on Goal Reasoning at Advances in Cognitive Systems (ACS2013), Baltimore, ML. 26-35.

Following the Radical Interactionism paradigm introduced previously, this paper investigates the construction of intentional actions and meaningful observations from regularities observed in sequences of sensorimotor interactions.

In doing so, we take the opposite stance of most machine learning approaches that seek to learn a mapping between a predefined space of actions and a predefined space of observations. Instead, our agent begins with a predefined space of interactions and learns actions and observations as secondary constructs. This paper also reports the Ernest 12 experiment.

Radical Interactionism

Olivier L. Georgeon and David W. Aha 2013. The Radical Interactionism Conceptual Commitment. Journal of Artificial General Intelligence 4(2): 31-36.

This is where I turn radical and eliminate perceptions and actions as primitive notions of cognitive models. Now I wonder why it took so many years to come up with such an obvious and elegant formalism.

Single Agents Can Be Constructivist Too

Olivier L. Georgeon and Salima Hassas 2013. Single Agents Can Be Constructivist too. Constructivist Foundations 9(1): 40-42.

In this open peer commentary to an article by Roesch et al.,  we wished to argue that multi-agent systems are not the only alternative to cognitivism. We present Ernest in Roesch et al.'s environment to show that Ernest is constructivist too.

Ernest 12

Ernest 12  categorizes the entities in its environment based on the possibilities of interaction that they afford, and adjusts its behavior to categories.

Top-left: Ernest in its environment. The "eye" (half-circle) takes the color of the entity that get Ernest's attention at any given time.

Top-right: Ernest's spatial memory. Interactions are localized in space, and Ernest updates their position as he moves. Entities are constructed where interactions overlap. Rectangles and trapezoids represent interactions, circles represent entities.

Bottom: activity trace. Bottom: interactions (rectangles and trapezoids) enacted on the left, in front, or on the right of Ernest. Middle: the motivational value of the enacted interaction represented as a bargraph (green when positive, red when negative). Top: the actions (half-circles (turn), triangles (try to step forward)) and the entities (blue and green circles) learned over time.

In this run, Ernest, learns the "bishop behavior" during the first 50 steps. On steps 78, we introduce two targets in a raw. The spatial memory shows that Ernest interacts with these two targets at the same time. Ernest's spatial memory (associated with its rudimentary attentional system) allows Ernest to focus on one target at a time.

On step 110, we introduce a "wall brick", and Ernest learns that this kind of entity affords the interaction "bumping". Subsequently, when we introduce a target, Ernest will preferably go towards the target than towards the wall brick because it has learned that the targets are edible.

Ernest 12 implements ECA, the Enactive Cognitive Architecture.
(Demo implemented with Ernest r439 and Vacuum r392)

Enactive Robot Learning

Olivier L. Georgeon, Christian Wolf, and Simon Gay 2013. An Enactive Approach to Autonomous Agent and Robot Learning.  IEEE Third Joint  International Conference on Development and Learning and on Epigenetic Robotics (EPIROB2013). Osaka, Japan. August 18-22 2013.

This paper constitutes a short introductory version of our ECA paper. It also presents the experiment of Ernest7 in an e-puck robot. 

Enactive Cognitive Architecture

Olivier L. Georgeon, James B. Marshall, and Riccardo Manzotti 2013. ECA: An enactivist cognitive architecture based on sensorimotor modeling. Biologically Inspired Cognitive Architectures, Volume 6, pp. 46-57, doi: 10.1016/j.bica.2013.05.006. Also presented at BICA2013.

This paper introduces a new way of modeling an agent interacting with an environment called an Enactive Markov Decision Process, inspired by the Theory of Enaction. It also describes Ernest's motivational principle in relation with the autotelic principle (Steels, 2004) and the optimal experience principle (Csikszentmihalyi, 1990). It introduces ECA, the Enactive Cognitive Architecture that drives Ernest 11, and it reports the Ernest 11.2 experiment.