Seminars

This talk presents frameworks for robust robotic manipulation, addressing the challenges of underactuated compliant hands and the estimation of Deformable Linear Objects (DLOs). First, we investigate the use of visuotactile perception to overcome the lack of proprioception in compliant hands during contact-rich insertion tasks. By employing a transformer-based policy trained through teacher-student distillation, we demonstrate that the fusion of high-resolution tactile feedback and extrinsic visual data enables successful, zero-shot sim-to-real transfer, ensuring precise task execution despite the inherent uncertainties of soft robotic hardware. Building on these tactile-sensing foundations, we extend our approach to the global shape estimation of DLOs. We introduce a physics-informed transformer architecture that maps dense, high-resolution tactile imprints to a 3D vertex chain, effectively resolving complex object configurations without the need for visual input. By enforcing structural priors derived from the Discrete Elastic Rods (DER) formulation, our model ensures physically consistent shape reconstruction even under occlusion. Collectively, these methods demonstrate that tactile-enabled perception is a critical modality for mastering both high-precision assembly and the dexterous manipulation of flexible materials.

Avishai Sintov is an Associate Professor at Tel Aviv University, where he leads the Robotics Lab within the School of Mechanical Engineering. His research focuses on human-robot interaction (HRI), dexterous in-hand manipulation, tactile sensing, and soft robotics. He is particularly engaged in developing machine learning models for robotics, including Diffusion Policies, Vision-Language-Action (VLA) models, and behavior cloning. Prof. Sintov holds a PhD in Mechanical Engineering from Ben-Gurion University and completed postdoctoral fellowships at the University of Illinois Urbana-Champaign and Rutgers University.