BEHAVIORS.AI is an Engine for enHancing verbAl and non-Verbal InteractiOns of RobotS, based on Artificial Intelligence – BEHAVIORS.AI

The LabCom BEHAVIORS.AI (Behaviors.ai is an Engine enHancing verbAl and non-Verbal InteractiOns of RobotS, based on Artificial Intelligence) aims at designing and developing an intelligent interaction engine to improve human-robot interactions and to make them more intuitive and more natural. This LabCom gathers researchers from the LIRIS research laboratory, and from Hoomano SAS. Hoomano makes software for social robots. BEHAVIORS.AI will enable us to transfer fundamental researches, notably in developmental learning to robotics applications dedicated to the society.
Many studies show that the main issue to improve interactions with machines is understanding natural spoken language. However, even if we observe significant improvements in this domain, the recognition error rate remains very problematic. Furthermore, the sole interpretation of a sentence, even if correct, is not always sufficient to understand the context of the interaction.
To develop better interactions, robots have to adapt themselves to the context (posture, tone, eyes-direction, behaviors, etc.) and have to be able to gradually improve their skills over time, interaction after interaction. BEHAVIORS.AI focuses both on improving context understanding, and on dynamically building adapted answers to situations.
Hence, the aim of the BEHAVIORS.AI project is to find ways to make interactions between humans and robots more empathic. This involves taking into account verbal and non-verbal communication (gestures, behaviors, facial expressions, feelings, looks, contextual information). We do not seek to improve a specific perception skill, but we want to combine existing inputs, whatever their quality, to better understand the context and to build more relevant and more adapted interactions, thus making the robot more adapted to its environment.

To reach this goal, we will develop an intelligent interaction intelligent engine able to use state of the art tools (speech recognition, faces, actions; deep learning; etc.), and we will augment it so that it can adapt better to new situations. Our researches are grounded on a very promising artificial intelligence technique which is developmental learning, and on cognitive robotics.

The research program of this LabCom is connected to the innovation strategy of Hoomano, which consists in producing generic tools so that they can be used on any kind of robotics platform. Our goal is to become a standard interaction engine on the market.

Besides the technical challenges related to the heterogeneity of the robotic platforms, this approach raises a major research challenge : the engine will have to be able to dynamically adapt to the available interactions on the robot, and to deal with incomplete or inconsistent information.


The agenda of BEHAVIORS.AI is based on short iterations allowing every version of the engine to be tested and used by users and clients. This approach is very important to get the necessary data and user feedbacks to go on with the research work, as well as for better understanding the users’ needs.