Global estimation of an object’s pose using tactile sensing

Abstract

It is essential for a successful completion of a robot object grasping and manipulation task to accurately sense the manipulated object's pose. Typically, computer vision is used to obtain this information, but it may not be available or be reliable in certain situations. This paper presents a global optimisation method where tactile and force sensing together with the robot's proprioceptive data are used to find an object's pose. This method is used to either improve an estimate of the object's pose given by vision or globally estimating it when no vision is available. Results show that the proposed method consistently improves an initial estimate (e.g. from vision) and is also able to find the object's pose without any prior estimate. To demonstrate the performance of the algorithm, an experiment is carried out where the robot is handed a small object (a pencil) and inserts it into a narrow hole without any use of vision.

Analytic grasp success prediction with tactile feedback

“…5. At present, we are not considering possibilities to reduce object pose uncertainty directly, such as using grasp execution schemes tailored to limit object movement [24] and/or to facilitate perception by incorporating tactile information [25]. Mechanism dynamics and control: While we exploit tactile feedback to aid in contact modeling as explained previously, wrench-based reasoning is blind to robot and controller dynamics (i. e., effects causing unforeseen movement of the grasp contacts due to the inability of computing/commanding appropriate grasp wrenches instantaneously) as stated in Section II.…”

Robotic tactile perception of object properties: A review

“…In [170] a Monte Carlo method was used to find an object pose where the local geometry of the object at the contact location matched the PCA features obtained from the tactile sensor. Another approach relied on the fact that, when the robot is in contact with the object, the possible locations of the object must lie inside a contact manifold, a novel particle filter was developed that was both faster and more accurate than a standard particle filter [171].…”

“…Based on the assumption that visually similar surfaces are likely to have similar haptic properties, vision is used to create dense haptic maps efficiently across visible surfaces with sparse haptic labels in [168]. Vision can also provide an approximate initial estimate of the object pose that is then refined by tactile sensing using local [173], [174] or global optimization [170].…”

“…It is concluded that combining vision and tactile sensing leads to the improvement of classification accuracy where one of them is weaker. In addition, some researchers have attempted to combine vision, tactile and force sensing to estimate the object pose in [170].…”

A survey of telerobotic surface finishing

“…Recently, Bimbo et al [20] have used global optimization by a genetic algorithm (GA) to estimate the pose of a given object by tactile sensing. Their application is gripping objects but a similar approach might be useful also in other robot operations like surface nishing.…”