Behaviors.ai

Behaviors.ai is a Common Laboratory funded by the ANR (French Research Agency) and involving people from Hoomano and from the SMA (Systèmes Multi-Agents) team of the LIRIS laboratory.  February 2017 – February 2020.

About

The Behaviors.ai joint laboratory will develop a smart interactional engine to provide more empathic, intuitive and natural human-robot interactions. This project gathers skills from LIRIS researchers in Artificial Intelligence and from Hoomano, a specialized firm in software development for social robots. Behaviors.ai targets to transfer fundamental research results, especially in developmental learning, to mass-market applications in robots.According to a lot of research studies, obtaining better natural spoken language recognition will drastically improve our interactions with machines. Even if significant progress is currently made in this domain, the recognition rates are still not sufficient. Moreover, the correct recognition of a sentence may still not be enough to characterize the context of the interaction. To obtain more empathic interactions, the robot has to dynamically understand the dialogue, which needs it to adapt its behavior (attitude, gaze orientation, type of behavior, …) to the context and to progressively learn new interactional schemas.

Team behaviors
People from Behaviors.ai lab and Hoomano
From left to right: Xavier Basset, Amélie Cordier, Pierre Laurent © David GAL-REGNIEZ

For these reasons, Behaviors.ai focuses on the understanding of the interactional context and on providing an appropriate and accurate response. The objectives of Behaviors.ai are to improve the context perception and the accuracy of the given response to provide more empathic human-robots interactions. This need to build mechanisms to interpret interactions based on verbal and non-verbal (gestures, attitudes, facial expressions, emotions, look, context) cues.

The user experience has thus to be rethought in this robotic context, which leads to changes in methodologies of conception to improve the human-robot coupling. This project does not target to improve a specific perceptual method but to propose smart ways to combine them to react appropriately, leading to more accurate and realistic interactions, reflecting a better adaptation of the robot to its environment.

This research project targets to develop an interactional engine that will include some of the state of the art methods and new innovative solutions, especially based on promising results in developmental learning and in cognitive robotics. It is part of the innovative development of Hoomano consisting in providing generic tools that can be deployed on any robotic platform to become a standard engine on the market. In addition to technical issues related to the diversity of robotic platforms, this objective raises a key research challenge: the engine has to be able to adapt dynamically to the interaction capabilities of the robot and to mature continuously depending on the effective and interesting interactions with its environment.

Partners

hoomano
label-ANR-bleu-CMJN
LIRIS et tutelles

Latest news

Follow us on twitter