The recent advancements in tactile and vision-based robotic sensing have significantly enhanced the capabilities of robots in complex environments. A notable trend is the integration of common-sense knowledge with sensory inputs, exemplified by frameworks like FusionSense, which enable robust 3D reconstruction from sparse views by fusing priors from foundation models with tactile and visual data. This approach addresses critical challenges in scene understanding and object manipulation, particularly for challenging objects such as those that are transparent or reflective. Another significant development is the introduction of active tactile sensors, such as DTactive, which combine tactile perception with in-hand manipulation, offering precision control during object interaction. These sensors leverage high-resolution tactile images and mechanical transmission mechanisms to achieve accurate angular trajectory control, demonstrating potential for robust in-hand manipulation tasks. Additionally, the field is witnessing innovations in tactile pattern reconstruction, with methods like TactileAR using low-resolution sensors to reconstruct high-resolution contact surfaces, enhancing the precision of robotic grasping and manipulation. These advancements collectively push the boundaries of robotic perception and manipulation, enabling more sophisticated and reliable interactions with the environment.

Noteworthy Papers:

• FusionSense revolutionizes sparse-view reconstruction by integrating common-sense priors with tactile and visual data, outperforming state-of-the-art methods.
• DTactive introduces an active surface to tactile sensors, enabling simultaneous tactile perception and in-hand manipulation with high precision.