A5057903433- Reinforcement Learning in Robotics
- Robot Manipulation and Learning
- AI-based Problem Solving and Planning
- Distributed and Parallel Computing Systems
- Micro and Nano Robotics
- Machine Learning and Algorithms
- Mathematical Dynamics and Fractals
- Modular Robots and Swarm Intelligence
- Robotics and Automated Systems
- Advanced Bandit Algorithms Research
- Research Data Management Practices
- Scientific Computing and Data Management
- Theoretical and Computational Physics
- Congenital heart defects research
- Educational Technology and Optimization
- Human Pose and Action Recognition
- Space Exploration and Technology
- Evolutionary Algorithms and Applications
- Action Observation and Synchronization
- Microbial Community Ecology and Physiology
- Embodied and Extended Cognition
- Neural dynamics and brain function
- Neurotransmitter Receptor Influence on Behavior
- Neural and Behavioral Psychology Studies
- Context-Aware Activity Recognition Systems
Okinawa Institute of Science and Technology Graduate University
2018-2022
Centre de Recherche Inria Bordeaux - Sud-Ouest
2016-2018
Institut national de recherche en informatique et en automatique
2013-2018
Institut Polytechnique de Bordeaux
2018
Centre National de la Recherche Scientifique
2018
Université de Bordeaux
2014-2018
Institut des Maladies Neurodégénératives
2018
Laboratoire Bordelais de Recherche en Informatique
2018
Synergy University
2017
École Nationale Supérieure de Techniques Avancées
2016
Computer science offers a large set of tools for prototyping, writing, running, testing, validating, sharing and reproducing results; however, computational lags behind. In the best case, authors may provide their source code as compressed archive they feel confident research is reproducible. But this not exactly true. James Buckheit David Donoho proposed more than two decades ago that an article about results advertising, scholarship. The actual scholarship full software environment, code,...
Scientific code is different from production software. code, by producing results that are then analyzed and interpreted, participates in the elaboration of scientific conclusions. This imposes specific constraints on often overlooked practice. We articulate, with a small example, five characteristics computational science should possess: re-runnable, repeatable, reproducible, reusable, replicable. The be executable (re-runnable) produce same result more than once (repeatable); it allow an...
The production of behavioral diversity---producing a diversity effects---is an essential strategy for robots exploring the world when facing situations where interaction possibilities are unknown or non-obvious. It allows to discover new aspects environment that cannot be inferred deduced from available knowledge. However, creating in it is most crucial---new and ones---is far trivial. In particular large redundant sensorimotor spaces, only small areas interesting explore any practical...
We show that goal-directed action planning and generation in a teleological framework can be formulated by extending the active inference framework. The proposed model, which is built on variational recurrent neural network characterized three essential features. These are (1) goals specified for both static sensory states, e.g., goal images to reached dynamic processes, moving around an object, (2) model cannot only generate plans, but also understand through observation, (3) generates...
We consider a scenario where an agent has multiple available strategies to explore unknown environment. For each new interaction with the environment, must select which exploration strategy use. provide strategy-agnostic method that treat situation as Multi-Armed Bandits problem reward signal is diversity of effects produces. test empirically on simulated planar robotic arm, and establish both able discriminate between dissimilar quality, even when differences are tenuous, resulting...
Evolution and development operate at different timescales; generations for the one, a lifetime other. These two processes, basis of much life on earth, interact in many non-trivial ways, but their temporal hierarchy-evolution overarching development-is observed most multicellular forms. When designing robots, however, this tenet lifts: It becomes-however natural-a design choice. We propose to inverse hierarchy developmental process happening phylogenetic timescale. Over classic evolutionary...
We present an algorithm for transferring exploration strategies between tasks that share a common motor space in the context of lifelong autonomous learning robotics. The does not transfer observations, or make assumptions about how is conducted. Instead, only selected commands are transferred tasks, chosen autonomously according to empirical measure progress. show on wide variety variations from source task, such as changing object robot interacting with altering morphology robot, this...
We propose the Reuse algorithm, that exploit data produced during exploration of an first environment to efficiently bootstrap second, different but related environment. The effect algorithm is produce a high diversity effects early exploration. only constrains environments share same motor space, and makes no assumptions about learning algorithms or sensory modalities. illustrate our on 6-joints robotic arm interacting with virtual object, show robust dissimilar environments, significantly...
We propose to make the physical characteristics of a robot oscillate while it learns improve its behavioral performance. consider quantities such as mass, actuator strength, and size that are usually fixed in robot, show when those at beginning learning process on simulated 2D soft performance locomotion task can be significantly improved. investigate dynamics phenomenon conclude our case, surprisingly, high-frequency oscillation with large amplitude for portion duration leads highest...
Evolution and development operate at different timescales; generations for the one, a lifetime other. These two processes, basis of much life on earth, interact in many non-trivial ways, but their temporal hierarchy -- evolution overarching is observed most multicellular lifeforms. When designing robots however, this tenet lifts: it becomes however natural design choice. We propose to inverse developmental process happening phylogenetic timescale. Over classic evolutionary search aimed...
We propose to make the physical characteristics of a robot oscillate while it learns improve its behavioral performance. consider quantities such as mass, actuator strength, and size that are usually fixed in robot, show when those at beginning learning process on simulated 2D soft performance locomotion task can be significantly improved. investigate dynamics phenomenon conclude our case, surprisingly, high-frequency oscillation with large amplitude for portion duration leads highest...