A5073978648- Reinforcement Learning in Robotics
- Evolutionary Algorithms and Applications
- Evolutionary Game Theory and Cooperation
- Evolution and Genetic Dynamics
- Neural dynamics and brain function
- Data Visualization and Analytics
- Domain Adaptation and Few-Shot Learning
- Data Stream Mining Techniques
- Robot Manipulation and Learning
- Machine Learning and Algorithms
- Social Robot Interaction and HRI
- Metaheuristic Optimization Algorithms Research
- Advanced Bandit Algorithms Research
- AI in Service Interactions
- Advanced Multi-Objective Optimization Algorithms
- Visual Attention and Saliency Detection
- Stochastic Gradient Optimization Techniques
- Advanced Image and Video Retrieval Techniques
- Robotic Path Planning Algorithms
- Robotic Locomotion and Control
- ICT in Developing Communities
- Prosthetics and Rehabilitation Robotics
- Gaussian Processes and Bayesian Inference
- EEG and Brain-Computer Interfaces
- Time Series Analysis and Forecasting
Sorbonne Université
2008-2025
Centre National de la Recherche Scientifique
2016-2025
Institut Systèmes Intelligents et de Robotique
2008-2023
Université Sorbonne Nouvelle
2018-2023
Imperial College London
2014-2015
Laboratoire Psychologie de la Perception
2012
Institut national de recherche en informatique et en automatique
2011
Laboratoire Jean Kuntzmann
2011
Université Grenoble Alpes
2011
Social robots have the potential to provide support in a number of practical domains, such as learning and behaviour change. This is particularly relevant for children, who proven receptive interactions with social robots. To reach therapeutic goals, issues need be investigated, notably design an effective child-robot interaction (cHRI) ensure child remains engaged relationship that educational goals are met. Typically, current cHRI research experiments focus on single type activity (e.g....
Reinforcement learning (RL) aims at building a policy that maximizes task-related reward within given domain. When the domain is known, i.e., when its states, actions and are defined, Markov Decision Processes (MDPs) provide convenient theoretical framework to formalize RL. But in an open-ended process, agent or robot must solve unbounded sequence of tasks not known advance corresponding MDPs cannot be built design time. This defines main challenges learning: how can learn behave...
Novelty Search is an exploration algorithm driven by the novelty of a behavior. The same individual evaluated at different generations has fitness values. corresponding landscape thus constantly changing and if, scale single generation, metaphor with peaks valleys still holds, this not case anymore whole evolutionary process. How does kind algorithms behave? Is it possible to define model that would help understand how works? This understanding critical analyse existing variants design new...
Abstract Quality-Diversity (QD) methods are algorithms that aim to generate a set of diverse and highperforming solutions given problem. Originally developed for evolutionary robotics, most QD studies conducted on limited domains'mainly applied locomotion, where the fitness behavior signal dense. Grasping is crucial task manipulation in robotics. Despite efforts many research communities, this yet be solved. cumulates unprecedented challenges literature: it suffers from reward sparsity,...
Performing Reinforcement Learning in sparse rewards settings, with very little prior knowledge, is a challenging problem since there no signal to properly guide the learning process. In such situations, good search strategy fundamental. At same time, not having adapt algorithm every single desirable. Here we introduce TAXONS, Task Agnostic eXploration of Outcome spaces through Novelty and Surprise algorithm. Based on population-based divergent-search approach, it learns set diverse policies...
As the physical limits of Moore's law are being reached, a research effort is launched to achieve further performance improvements by exploring computation paradigms departing from standard approaches. The BAMBI project (Bottom-up Approaches Machines dedicated Bayesian Inference) aims at developing hardware probabilistic computation, which extends logic realised boolean gates in current computer chips. Such computing devices would allow solve faster and lower energy cost wide range...
We present a new method to visualize uncertain scalar data fields by combining color scale visualization techniques with animated, perceptually adapted Perlin noise. The parameters of the noise are controlled uncertainty information produce animated patterns showing local value and quality. In order precisely control perception patterns, we perform psychophysical evaluation contrast sensitivity thresholds for set stimuli. validate extend this using an existing computational model. This...
Evolvability is an important feature that impacts the ability of evolutionary processes to find interesting novel solutions and deal with changing conditions problem solve. The estimation evolvability not straight-forward generally too expensive be directly used as selective pressure in process. Indirectly promoting a side effect other easier faster compute selection pressures would thus advantageous. In unbounded behavior space, it has already been shown evolvable individuals naturally...
In the past few years, a considerable amount of research has been dedicated to exploitation previous learning experiences and design Few-shot Meta Learning approaches, in problem domains ranging from Computer Vision Reinforcement based control. A notable exception, where best our knowledge, little no effort made this direction is Quality-Diversity (QD) optimization. QD methods have shown be effective tools dealing with deceptive minima sparse rewards Learning. However, they remain costly due...
Reward-based optimization algorithms require both exploration, to find rewards, and exploitation, maximize performance. The need for efficient exploration is even more significant in sparse reward settings, which performance feedback given sparingly, thus rendering it unsuitable guiding the search process. In this work, we introduce SparsE Reward Exploration via Novelty Emitters (SERENE) algorithm, capable of efficiently exploring a space, as well optimizing rewards found potentially...
Novelty search was proposed as a means of circumventing deception and providing selective pressure towards novel behaviours to provide path open-ended evolution.Initial implementations relied on neuro-evolution approaches which increased network complexity over time.However, although many studies have reported impressive results, it is still not clear whether the benefits evolving topologies are outweighed by overall approach.Given that novelty can also be combined with evolutionary methods...
Robots are still limited to controlled conditions, that the robot designer knows with enough details endow appropriate models or behaviors. Learning algorithms add some flexibility ability discover behavior given either demonstrations a reward guide its exploration reinforcement learning algorithm. Reinforcement rely on definition of state and action spaces define reachable Their adaptation capability critically depends representations these spaces: small discrete result in fast while large...
We report first results on children adaptive behavior towards a dance tutoring robot. can observe that rapidly evolves through few sessions in order to accommodate with the robotic tutor rhythm and instructions.
Learning optimal policies in sparse rewards settings is difficult as the learning agent has little to no feedback on quality of its actions. In these situations, a good strategy focus exploration, hopefully leading discovery reward signal improve on. A algorithm capable dealing with this kind setting be able (1) explore possible behaviors and (2) exploit any discovered reward. Exploration algorithms have been proposed that require definition low-dimension behavior space, which generated by...
Evolution in nature has allowed biological systems to develop and survive many different environments. It can be attributed one of the major features natural evolution: its open-endedness, that considered as ability continuously produce novelty and/or complexity [1]. This feature is critical allow an agent adapt environment. We propose here extension Quality Diversity algorithms make it more open-ended. aim at generating a large set diverse solutions [3, 10]. They used two-steps process...
As open-ended learning based on divergent search algorithms such as Novelty Search (NS) draws more and attention from the research community, it is natural to expect that its application increasingly complex real-world problems will require exploration operate in higher dimensional Behavior Spaces (BSs) which not necessarily be Euclidean. traditionally relies k-nearest neighbours an archive of previously visited behavior descriptors are assumed live a Euclidean space. This problematic...
To solve its task, a robot needs to have the ability interpret perceptions. In vision, this interpretation is particularly difficult and relies on understanding of structure scene, at least extent task sensorimotor abilities. A with build adapt process according own tasks capabilities would push away limits what robots can achieve in non controlled environment. solution provide processes such representations that are not specific an environment or situation. lot works focus objects...
Robotic grasping refers to making a robotic system pick an object by applying forces and torques on its surface. Despite the recent advances in data-driven approaches, still needs be solved. In this work, we consider as Diversity Search problem, where attempt find many solutions possible that verify sparse binary criterion. We propose variant of state-of-the-art QD method for based divide-and-conquer paradigm handle discontinuities. Experiments conducted 3 different robot-gripper setups...
Quality-Diversity (QD) methods are algorithms that aim to generate a set of diverse and high-performing solutions given problem. Originally developed for evolutionary robotics, most QD studies conducted on limited domains - mainly applied locomotion, where the fitness behavior signal dense. Grasping is crucial task manipulation in robotics. Despite efforts many research communities, this yet be solved. cumulates unprecedented challenges literature: it suffers from reward sparsity, behavioral...
Robots need to understand their environment perform task. If it is possible pre-program a visual scene analysis process in closed environments, robots operating an open would benefit from the ability learn through interaction with environment. This furthermore opens way acquisition of affordances maps which action capabilities robot structure its understanding. We propose approach build such by relying on interactive perception and online classification. In proposed formalization...
Evolvability is an important feature that impacts the ability of evolutionary processes to find interesting novel solutions and deal with changing conditions problem solve. The estimation evolvability not straightforward generally too expensive be directly used as selective pressure in process. Indirectly promoting a side effect other easier faster compute selection pressures would thus advantageous. In unbounded behavior space, it has already been shown evolvable individuals naturally...