Phd Position F/m Anticipating Human Behavior for

Le descriptif de l’offre ci-dessous est en Anglais_ **Type de contrat **:CDD **Niveau de diplôme exigé **:Thèse ou équivalent **Fonction **:Doctorant **A propos du centre ou de la direction fonctionnelle**: The Inria Grenoble research center groups together almost 600 people in 23 research teams and 7 research support departments. Staff is present on three campuses in Grenoble, in close collaboration with other research and higher education institutions (University Grenoble Alpes, CNRS, CEA, INRAE,), but also with key economic players in the area. Inria Grenoble is active in the fields of high-performance computing, verification and embedded systems, modeling of the environment at multiple levels, and data science and artificial intelligence. The center is a top-level scientific institute with an extensive network of international collaborations in Europe and the rest of the world. **Contexte et atouts du poste**: **Collaboration between Inria team THOTH and the multidisciplinary institute in artificial intelligence (MIAI) in Grenoble** The close connection to University of Grenoble Alpes offers additionally many oportunities for collaboration, further training and networking within and cross research fields. **Mission confiée**: **Keywords** human-robot interaction, machine learning, computer vision, representation learning **Theme/Domain** Human-robot systems are challenging because the actions of one agent can significantly influence the actions of others. Therefore, anticipating the partner's actions is crucial. By inferring beliefs, intentions, and desires, we can develop cooperative robots that learn to assist humans or other robots effectively. In this project we are in particular interested in estimating human intentions to enable collaborative tasks between humans and robots such as human-to-robot and robot-to-human handovers. **Context and Motivation** Even after three decades of research in human-robot interaction, current systems still struggle to comprehend users’ intentions, goals, and needs, thus failing to anticipate their actions. A shared "non-verbal" language is missing, which would enable communication beyond classical user commands for robot control. This absence of a common representation of behavioral skills, which would allow anticipation of each other's behavior, limits an agent's ability to interact and collaborate effectively with humans. We aim to build a new generation of human-robot interaction that pro-actively adapts to users’ future input actions by monitoring their movements and predicting their interaction intentions. **Contact** The thesis will be jointly supervised by Pia Bideau (THOTH), Karteek Alahari (THOTH) and Xavier Alameda Pineda (RobotLearn). **Principales activités**: **Summary** This project aims to estimate human intentions within a collaborative human-robot setup using visual data. Traditional approaches to action classification focus on identifying and categorizing immediate actions. However, collaborative tasks, such as human-to-robot handovers, demand a more nuanced understanding. These tasks require not only the ability to classify for example the type of the human grasp accurately but also to anticipate future actions effectively. To achieve this, the project will: - Identify Human Hand-Object Configurations: Develop algorithms to precisely identify and classify the positioning and interaction of human hands with objects. This involves understanding the object's shape and its impact on the human grasp pose. - Anticipate Future Actions: Implement predictive models that can estimate what actions a human might take next. This anticipation is crucial for seamless collaboration, allowing robots to prepare and respond proactively to human movements. By combining these elements, the project seeks to create a more intuitive and effective human-robot interaction framework. This will enable robots to better understand and predict human behavior, leading to smoother and more intuitive collaborations for a variety of handover tasks. **Goals** - Developing Latent Action Representations: Utilize generative models to create latent action representations that can handle both classification and anticipation tasks. These representations will capture the underlying structure of human actions, allowing the system to recognize and predict behaviors more effectively. - Addressing Challenges of Low Data Availability: Implement strategies to work effectively in environments with limited data. This may involve techniques such as transfer learning, data augmentation, and synthetic data generation to ensure robust performance despite scarce training examples. - Managing Initially Unknown Classes: Develop methods to identify and incorporate new, previously unseen classes of actions. This capability will enable the system to adapt and learn in real-time, enhancing its ability to function in dynamic and unpredictable environments. By achieving these obj