RHUM

Scientific Description

Robots offer a unique way to bridge the virtual and physical worlds by enriching the user experience with digital content through contextual processing of the physical scene, and by acting in the physical world as digital agents.

The RHUM action team explores new forms of active perception of the physical environment and will create adaptive forms of social communication and interaction between humans and robots, considered as digital agents in the physical world. The learning of human- and situation-sensitive behaviors will also be studied through the online and active exploration of multimodal data streams.

Main results and future work

Learning and adapting robot behaviors for social interaction

Equipping robots with socio-communicative skills to engage them in effective human-robot interaction (HRI) is a major challenge. We are exploring several paradigms for training multimodal behavioral models from human demonstrations: (a) learning by observing human-robot interactions and adapting sensorimotor events to the capabilities of the target robot [1]; (b) learning from behaviors collected via immersive teleoperation of the robot by human pilots [2]. Particular attention is also paid to the evaluation of these HRI behaviors, notably through original online evaluation paradigms [3].

Omar Samir Mohammed's thesis, "Acquiring Humain Robot Interaction skills with transfer learning techniques", supervised by Gérard Bailly and Damien Pellier, will focus on acquiring new interactive skills through the transfer of previous experiences or skills already mastered by another robot. We will also study the decomposition of tasks into generic sensorimotor skills and interactive skills. The first application will focus on the transfer learning of cursive handwriting from drawings (i.e., captioning) between GIPSA-Lab's iCub and LIG's Baxter.

Human-Robot Motion (HRM) 

HRM consists of controlling how a robot moves among people. In this case, it is essential that the robot's movements are both safe and appropriate. Safe means that the robot must not injure people by colliding with them, and appropriate means that its movements must respect the social and cultural norms that govern people's behavior, for example by avoiding walking between two people who are talking.

The work carried out at RHUM has focused primarily on safety. Avoiding collisions with people is challenging because their future behavior is uncertain and difficult to anticipate. This rules out any "absolute" safety, meaning collisions may be unavoidable in certain circumstances. Our goal is therefore to study the level of safety achievable in a given situation. Considering a bipedal robot and using tools previously developed within the team, we proposed a strategy allowing a bipedal robot to move through a crowd while maintaining balance and ensuring passive safety, meaning that in the event of a collision, the robot will come to rest [4]. This work was further developed during Hang Yu's master's internship, funded by RHUM [5], which studied a more sophisticated level of safety called passive friendly safety—meaning that in the event of a collision with a person, the robot will come to rest and the person will be able to avoid the collision if they choose to. Matteo Ciocca's PhD, funded by RHUM and supervised by Thierry Fraichard and Pierre-Brice Wieber, will build upon and expand the work carried out in [5].

Coordinators

Gérard Bailly (GIPSA-Lab)

Pierre-Brice Wieber (Inria/LJK)

Olivier Aycard (LIG)

Technology transfer

Several projects complement this action team:

• PSPC 1 ROMEO 2 "Development of an assistant and companion humanoid robot for daily life" (2012-2017, LJK-BIPOP) with ALDEBARAN
• H2020 COMANOID "Multi-Contact Collaborative Humanoids in Aircraft Manufacturing" (2015-2019, LJK-BIPOP) with AIRBUS
• ANR SOMBRERO "Immersive teleoperation of humanoid robots and learning of socio-communicative behavior models" (2014-2018, GIPSA-CRISSP, LIG-MAGMA, LIP, and Lab-STICC) with ALDEBARAN
• Rhône-Alpes Region ARC6 TENSIVE Project "Telepresence robots: Immersive Social Navigation and Verbal Interaction" (2016-2019, GIPSA-CRISSP and CITI-CHROMA) with AWABOT and HUMANOO

Significant Publications

[1] A. Mihoub, G. Bailly, C. Wolf and F. Elisei, "Graphical models for social behavior modeling in face-to face interaction", Pattern Recognition Letters, vol. 74, 2016.

[2] G. Gomez, C. Plasson, F. Elisei, F. Noël, and G. Bailly, "Qualitative assesment of a beaming environment for collaborative professional activities" in European Conf. for Virtual Reality and Augmented Reality (EuroVR), 2015.

[3] N. Duc-Canh, G. Bailly and F. Elisei, "Conducting neuropsychological tests with a humanoid robot: design and evaluation" in IEEE Int. Conf. on Cognitive InfoCommunications (CogInfoCom), 2016.

[4] N. Bohórquez, A. Sherikov, D. Dimitrov, and P.-B. Wieber, "Safe navigation strategies for a biped robot walking
in a crowd" in IEEE-RAS Int. Conf. on Humanoid Robots (Humanoids), 2016.

[5] H. Yu, "Safe Navigation of Biped Robots Subject to Passive Friendly Safety and Balance Constraints", Master Report, Univ. Grenoble-Alpes, 2016.