Embedded Files
University of Évry Paris-Saclay, Building Maupertuis
Overview
Robotic systems are moving beyond structured industrial environments into open, human-centered, and physically dynamic spaces. To operate reliably in these settings, next-generation robots must do more than perceive, plan, or control in isolation. They must integrate cognition, multimodal sensing, and embodied action into a unified intelligence stack. This workshop focuses on the convergence of three pillars of future robotics: cognitive intelligence for reasoning, planning, memory, and decision-making; sensing intelligence for multimodal perception, proprioception, tactile understanding, and real-time scene interpretation; and embodied intelligence for learning, control, interaction, and physical adaptation. By bringing together researchers working on robot learning, perception, control, human–robot interaction, foundation models, embodied AI, and robotic system integration, the workshop aims to identify the core principles, architectures, and evaluation methods needed to build robots that can understand the world, act through their bodies, and adapt safely in real-world environments.
Invited Speakers:
INALCO, France
Topic:
Cultures and Values: From Large Language Models to Robotic Systems
This talk examines the place accorded to the notion of culture in industry and academic work on the design of large language models (LLMs). A first observation is that culture is approached therein in a restrictive manner, reduced to questions of moral value. A second observation is that current work on the cultural dimension of LLMs falls into two broad categories: (i) the evaluation of cultural bias through confrontation with third-party cultural benchmarks, and (ii) values-based alignment of models. We will discuss the implications of these anthropological assumptions for the future development of artificial intelligence and robotic systems.
ISIR, Sorbonne University, France
Topic:
Real-time Intention Prediction from Partial Video Streams
This talk addresses real-time intention prediction—anticipating a person's future actions before they unfold. Unlike conventional action recognition, intention prediction infers latent, unobservable constructs from incomplete data, which is crucial for proactive systems like assistive robotics and autonomous vehicles. The presented approach processes partial video streams incrementally to output a probability distribution over potential intentions. Moving beyond fixed, predefined categories, this method operates in an open-vocabulary setting, enabling zero-shot prediction on unseen intentions. The talk concludes with experimental results across datasets and future research directions.
Larbi Yousfi
ISAE-SUPMECA, France
University of Sfax, Tunisia
Topic:
Physics-Informed AI for Smart Robotic–Mechatronic Additive Manufacturing
Traditional finite-element simulations are far too slow for real-time parameter adjustment in directed energy deposition (DED) 3D printing, which is why we developed a fast, physics-constrained 3D physics-informed neural network (PINN) coupled with ONNX Runtime to embed thermodynamic laws directly into the AI. When tested on multi-layer Ti-6Al-4V DED, our framework successfully retained the expected melt-pool morphology and predicted peak temperatures with a mere 1.14% deviation from literature benchmarks, all while maintaining an incredibly low global RMSE during PyTorch-to-ONNX verification. Crucially, the model requires no retraining and generates high-resolution thermal maps in just 54 seconds per layer, shifting thermal simulation from slow offline analysis to rapid parameter screening that will enable future sensor-informed, real-time robotic DED control.
Hunan University, China
Topic:
Biomechanics-Informed Embodied Intelligence for Personalized Wearable and Rehabilitation Robotics
Recent advances in embodied AI have improved robotic perception and decision-making, yet safe human–robot interaction remains challenging due to the complexity of the human neuromusculoskeletal system. This talk presents a biomechanics-informed embodied intelligence framework for personalized wearable and rehabilitation robotics. By integrating multimodal sensing, neuromusculoskeletal modeling, physics-informed learning, and digital twins, the framework enables subject-specific modeling, robust prediction of human motion and biomechanics, estimation of unmeasurable variables, and safety evaluation before deployment. Recent advances in markerless upper-limb assessment, multimodal wearable sensing, and brain–body collaborative intelligence are also presented. The proposed framework demonstrates how biomechanics and embodied intelligence can enable safer, more adaptive, and clinically meaningful robotic systems for healthcare, rehabilitation, and human augmentation.
Organizers:
Workshop Program
We invite submissions presenting original research, late-breaking results, work-in-progress studies, system demonstrations, and perspective papers related to cognitive, sensing, and embodied intelligence for robotics. The workshop is intended as a high-impact venue for open discussion, technical exchange, and forward-looking debate. Submissions will be evaluated based on technical quality, relevance to the workshop theme, clarity of presentation, and potential to stimulate discussion.
Topics of Interest
Topics of interest include, but are not limited to:
Cognitive Intelligence for Robotics
Cognitive architectures for autonomous and interactive robots
Robot reasoning, task planning, memory, and decision-making
Foundation models, world models, vision-language-action models, and robot policy models
Knowledge representation, semantic mapping, and scene understanding
Uncertainty-aware, risk-aware, and context-aware robot cognition
Continual learning, lifelong adaptation, and robot self-improvement
Sensing Intelligence and Multimodal Perception
Multimodal sensor fusion for robotics
Visual, tactile, auditory, force, torque, inertial, and proprioceptive perception
Active perception and sensorimotor exploration
Tactile intelligence, haptic perception, and contact understanding
Event-based sensing, neuromorphic perception, and low-latency sensing pipelines
Perception under uncertainty, occlusion, clutter, deformation, and dynamic environments
Embodied Learning, Control, and Interaction
Embodied AI and sensorimotor learning
Imitation learning, reinforcement learning, and self-supervised learning for robots
Whole-body control, dexterous manipulation, locomotion, and mobile manipulation
Contact-rich manipulation and physical interaction
Morphology-aware intelligence and hardware–software co-design
Sim-to-real transfer, digital twins, and real-world policy deployment
Human-Centered and Safe Robotic Systems
Human–robot interaction and collaboration
Learning from human feedback, demonstrations, and shared autonomy
Assistive, wearable, rehabilitation, and healthcare robotics
Safety-aware embodied decision-making and interaction
Explainable and trustworthy embodied robot behavior
Ethical, social, and usability considerations in next-generation robot systems
Invited Speakers
Invited Speakers
Invited speakers are asked to provide the following information for the workshop program:
Talk title
Short abstract, 100–200 words
Short biography, 100–150 words
Speaker photo
Affiliation and optional personal website link
The information should be sent to the email address hulxhlx@gmail.com. A full paper is not required for invited speakers.
Important Dates
Workshop: July 20th 2026
ICARM Conference: July 21st - 23rd 2026
Page updated
Google Sites
Report abuse