AI for improving machine interactions with humans and the environment

This research event aims at bridging the gap between researchers in computer science and robotics both working on the problem of human / agent interaction using artificial intelligence tools. The event is supported by the LIRMM laboratory and generally aims at contributing to cross-domain scientific collaborations within the laboratory.

27th October

13:00-14:00 Welcome coffee

13:50-14:00 Overview

14:00-14:45 Rachid Alami: Models and Decisional issues for Human-Robot Joint Action

This talk will address some key decisional issues that are necessary for a cognitive and interactive robot which shares space and tasks with humans. We adopt a constructive approach based on the identification and the effective implementation of individual and collaborative skills. The system is comprehensive since it aims at dealing with a complete set of abilities articulated so that the robot controller is effectively able to conduct in a flexible and fluent manner a human-robot joint action seen as a collaborative problem solving and task achievement. These abilities include geometric reasoning and situation assessment based essentially on perspective-taking and affordances, management and exploitation of each agent (human and robot) knowledge in a separate cognitive model, human-aware task planning and interleaved execution of shared plans. We will also discuss the key issues linked to the pertinence and the acceptability by the human of the robot behaviour, and how this influence qualitatively the robot decisional, planning, control and communication processes.

14:45-15:30 Judith Masthoff: Artificial Intelligence for Emotional Wellbeing

Abstract: Researchers claim that we are facing a global loneliness epidemic, and that mental illness, anxiety disorders, stress and burnout are on the rise. This talk is about how adaptive AI systems can actively improve emotional wellbeing. We will discuss different ways of doing so, the work already done, the challenges faced, and our vision of a new kind of personalized systems. First, systems can provide emotional support, adapted to the recipient's characteristics such as their personality, affective state, cultural background, and stressors experienced. Second, systems can aid humans to provide emotional support, so mediating emotional support, adapting to both the support giver and recipient. Third, systems can support and motivate people to adopt behaviours that improve their well-being and that of others. Fourth, systems can team people up, deciding who are the best placed to provide support and motivation. Finally, systems can improve the well-being of groups and not just individuals, monitoring group wellbeing, encouraging and supporting effective group behaviours, and building group identity and cohesion.

15:30-15:45 Coffee

15:45 - 16:30 Serena Ivaldi : Using machine learning to optimize whole-body teleoperation of humanoid robots

Humanoid robots have a potential to find applications as human co-workers or human avatars, and recent advances in mechatronics of bipeds make this goal closer than ever. In both applications, humanoids need to exhibit complex movement skills, from locomotion to whole-body manipulation. We recently proposed to use teleoperation to demonstrate such complex skills, which fully leverage the intelligence of the human operator to make real-time decisions about the activities to do and how to do them. However, a robot controller whole-body teleoperation needs three critical features: robustness, to prevent the robot to fall; generality, to be able to execute in principle any possible task commanded by the operator; predictability, to anticipate the operator’s intention in case of communication delays. In this talk I will present our work combining machine learning and control to address these three features, in particular how we used probabilistic models and multi-objective optimization to find a practical solution to each issue. I will present results on both the iCub and the Talos humanoid robots

16:30-17:30 Panel Discussion

28th October

09:30-09:45 Coffee and croissants

09:45-10:30 Patrick van der Smagt

Control of multidimensional systems typically relies on accurately engineered models. Breaking this requirement is problematic with neural networks, as their Gauss-distributed data assumptions typically do not hold. We address this problem with sequential latent-variable models. Trick: regularisation of the latent space using model or control knowledge. Take-away: it works, as we demonstrate on various interactions through drones, robots, or drummers.

10:30-11:15 Jean-Baptiste Mouret: Data-efficient learning for fast damage recovery in robotics

The recent advances in deep learning are generating an impressive interest in machine learning, but their influence on robotics is not as strong as we could think (yet). In this talk, I will show in which situations robots can benefit the most from learning and what constraint robots impose on learning algorithms. Focusing on trial-and-error learning, I will then introduce the work of our team to address these challenges, in particular to allow legged robots to recover from mechanical damage in a few minutes.

11:15 - 12:00 LIRMM visit

12:00-14:00 Lunch