Differences
This shows you the differences between two versions of the page.
Both sides previous revision Previous revision | |||
members:sturmju:research:tactile [2011/07/12 11:27] sturmju |
members:sturmju:research:tactile [2012/01/20 12:01] (current) |
||
---|---|---|---|
Line 1: | Line 1: | ||
+ | ====== Tactile Sensing ====== | ||
+ | |||
+ | In particular for robotic manipulation | ||
+ | tasks, tactile sensing provides another sensor modality that can | ||
+ | reveal relevant aspects about the object being manipulated, | ||
+ | example, to infer its identity, pose, and internal state. | ||
+ | |||
+ | ==== Tactile object recognition using the bag-of-features approach ==== | ||
+ | |||
+ | Our approach uses the bag-of-features model that we apply to object | ||
+ | classification based on tactile images. First, the robot generates a | ||
+ | suitable tactile feature vocabulary using unsupervised clustering from | ||
+ | real data. Second, it learns a set of feature models (a so-called | ||
+ | codebook that encodes the appearance of objects in form of | ||
+ | feature histograms. After training, a robot can use this codebook to | ||
+ | identify the grasped object. Since the objects that we | ||
+ | consider are typically larger than the sensor and consist of similar | ||
+ | parts, the robot may need multiple grasps at different | ||
+ | positions to uniquely identify an object. | ||
+ | number of required grasps, we apply a decision-theoretic framework that chooses | ||
+ | the grasping location in such a way that the expected entropy of the belief | ||
+ | distribution is minimized. | ||
+ | and household objects, we demonstrate that our approach enables a | ||
+ | manipulation robot to discriminate various objects only by touch. | ||
+ | |||
+ | |||
+ | < | ||
+ | <iframe width=" | ||
+ | <iframe width=" | ||
+ | </ | ||
+ | |||
+ | ==== Estimating the internal state of containers ==== | ||
+ | We investigated features that describe the | ||
+ | dynamics in the tactile response while the robot is | ||
+ | grasping or manipulating an object. As we showed, the dynamic | ||
+ | components of the tactile signal can be used to infer several aspects of the internal state | ||
+ | of an object. For example, these features allow a robot to | ||
+ | detect whether a grasped bottle contains liquid and whether its cap | ||
+ | has been properly closed. This information is highly relevant for | ||
+ | domestic robots that fulfill service tasks such as tidying up. | ||
+ | |||
+ | |||
+ | |||
+ | ====== Related Publications ====== | ||
+ | |||
+ | < | ||
+ | < | ||
+ | < | ||
+ | </ | ||