Projektdetaljer
Beskrivelse
The aim of this project is to enable robots to perceive the pose and geometry of objects by tactile sensing, like humans use their sense of touch. Current robots are challenged by lack of tactile perception, as they are often only equipped with cameras and simple force sensors. This is a barrier to accurate robotic manipulation, especially for small and flexible objects. The robotics community is currently putting a lot of resources into the development of tactile sensors that can cover surfaces such as robot hands. Such sensors open new possibilities for the use of robotics for manipulation of small and flexible objects like assembly of delicate products – this will be our target application.
The research questions that will be addressed in this project are
• How can the pose of a known object be determined from tactile information?
• How can the grasp of an object be altered to decrease the pose uncertainty of a known object?
Tactile perception algorithms are often inspired by computer vision algorithms; however, the relation between finger positions and object must be integrated into tactile perception algorithms to obtain accurate measurements. Also, tactile sensors measure both forces tangential and normal to the object surface; this complicates the estimation problem. Our aim is to develop a tactile perception algorithm with sub-millimeter accuracy for objects with known geometry, by exploiting knowledge about finger geometry.
A robot can change its grasp of an object by either regrasping, using extrinsic dexterity, or in-hand manipulation. The possibility to reorient a grasped object makes it possible to obtain new tactile information that can be used for determining the pose of an object. This is called active tactile perception. We aim to develop an active tactile perception algorithm, based on a manifold particle filter, that determines a regrasp that minimizes pose uncertain while respecting physical constraints of the robot system.
The project is funded by Fabrikant Vilhelm Pedersen og Hustrus Legat.
The research questions that will be addressed in this project are
• How can the pose of a known object be determined from tactile information?
• How can the grasp of an object be altered to decrease the pose uncertainty of a known object?
Tactile perception algorithms are often inspired by computer vision algorithms; however, the relation between finger positions and object must be integrated into tactile perception algorithms to obtain accurate measurements. Also, tactile sensors measure both forces tangential and normal to the object surface; this complicates the estimation problem. Our aim is to develop a tactile perception algorithm with sub-millimeter accuracy for objects with known geometry, by exploiting knowledge about finger geometry.
A robot can change its grasp of an object by either regrasping, using extrinsic dexterity, or in-hand manipulation. The possibility to reorient a grasped object makes it possible to obtain new tactile information that can be used for determining the pose of an object. This is called active tactile perception. We aim to develop an active tactile perception algorithm, based on a manifold particle filter, that determines a regrasp that minimizes pose uncertain while respecting physical constraints of the robot system.
The project is funded by Fabrikant Vilhelm Pedersen og Hustrus Legat.
| Akronym | VP-Tactile Perception |
|---|---|
| Status | Igangværende |
| Effektiv start/slut dato | 01/01/2026 → 31/12/2026 |