PhD Position: Multimodal Perception and Hybrid Control for Robotic Manipulation

This PhD position focuses on advancing robotic manipulation in unstructured environments by integrating multimodal perception and hybrid control strategies. The project, embedded within the HAMMER initiative under the PEPR Robotique program, aims to enable robots to handle deformable, flexible, or partially occluded objects—such as textiles—by combining geometric, physical, and learned representations. The candidate will design frameworks that adaptively reason about object states, contact dynamics, and task affordances, bridging model-based constraints with data-driven flexibility. Work will be conducted on a collaborative robot (cobot) equipped with diverse sensory modalities, with experimental validation in real-world scenarios. The research aligns with cutting-edge trends in foundation models, vision-language-action systems, and safety-aware control, addressing open challenges in robotic autonomy for dynamic environments.

• Develop multimodal representations combining RGB-D geometry, local shape descriptors, contact indices, proprioceptive signals, and learned features for high-dimensional object states.
• Design hybrid control strategies that integrate model-based physical constraints with data-driven adaptation to handle uncertainty in object properties and interactions.
• Anchore high-level action priors (e.g., from foundation models) in real-time physical feedback to refine robotic manipulation tasks.
• Implement and validate proposed frameworks on a cobot equipped with diverse sensory modalities, focusing on real-world scenarios with deformable/articulated objects.
• Publish findings in top-tier robotics conferences and journals, contributing to the broader HAMMER project goals under PEPR Robotique.
• Collaborate with interdisciplinary teams to integrate perception, control, and learning components into a cohesive robotic manipulation system.