SaddleSense: Saddle-Shaped Modules for 3D Shape Sensing with Single-Axis Off-the-Shelf Bend Sensors
SaddleSense: Saddle-Shaped Modules for 3D Shape Sensing with Single-Axis Off-the-Shelf Bend Sensors
This research introduces SaddleSense, a 3D-printed TPU saddle-shaped lattice that integrates single-axis off-the-shelf bend sensors to capture multi-axis deformation in continuum robots. The system uses a hybrid calibration combining learned models with physics-based fallbacks to map sensor readings to piecewise constant curvature parameters. A six-module prototype on a 430 mm body achieves a tip position RMSE of 13.2 mm (3.1%) on held-out samples similar to the training set and 15.8 mm (3.7%) on separate trials, demonstrating effective posture feedback without tying segment length to sensor length.
The study presents a novel saddle-shaped lattice design using off-the-shelf bend sensors to enable two-axis curvature sensing for continuum robots, achieving a tip position RMSE of 13.2 mm on held-out samples similar to the training set and 15.8 mm on separate trials with a six-module prototype.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
tip position RMSE
29.9
View original evidence
bration maps the sensor readings to piecewise constant curvature (PCC) parameters using a learned model paired with a physics-based fallback for out-of-distribution inputs. We demonstrate a six-module prototype with 24 off-the-shelf bend sensors on a 430~mm continuum body. Using only each module's four local readings, the physics-based model alone achieves a tip position root mean square error (RMSE) of 29.9~mm (7.0\
Open source S4tip position RMSE
13.2
View original evidence
% of body length). The data-driven model reduces the tip RMSE to 13.2~mm (3.1\%) on held-out samples of motions similar to the training set and 15.8~mm (3.7\%) on separately recorded trials. Reducing each module from four sensors to one nearly triples the tip error.
Open source S5tip position RMSE
15.8
View original evidence
% of body length). The data-driven model reduces the tip RMSE to 13.2~mm (3.1\%) on held-out samples of motions similar to the training set and 15.8~mm (3.7\%) on separately recorded trials. Reducing each module from four sensors to one nearly triples the tip error.
Open source S5body length
430
View original evidence
bration maps the sensor readings to piecewise constant curvature (PCC) parameters using a learned model paired with a physics-based fallback for out-of-distribution inputs. We demonstrate a six-module prototype with 24 off-the-shelf bend sensors on a 430~mm continuum body. Using only each module's four local readings, the physics-based model alone achieves a tip position root mean square error (RMSE) of 29.9~mm (7.0\
Open source S4modules
6
View original evidence
bration maps the sensor readings to piecewise constant curvature (PCC) parameters using a learned model paired with a physics-based fallback for out-of-distribution inputs. We demonstrate a six-module prototype with 24 off-the-shelf bend sensors on a 430~mm continuum body. Using only each module's four local readings, the physics-based model alone achieves a tip position root mean square error (RMSE) of 29.9~mm (7.0\
Open source S4
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
bration maps the sensor readings to piecewise constant curvature (PCC) parameters using a learned model paired with a physics-based fallback for out-of-distribution inputs. We demonstrate a six-module prototype with 24 off-the-shelf bend sensors on a 430~mm continuum body. Using only each module's four local readings, the physics-based model alone achieves a tip position root mean square error (RMSE) of 29.9~mm (7.0\
Open source S4
% of body length). The data-driven model reduces the tip RMSE to 13.2~mm (3.1\%) on held-out samples of motions similar to the training set and 15.8~mm (3.7\%) on separately recorded trials. Reducing each module from four sensors to one nearly triples the tip error.
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org