Organizations: All authors are with the Human Interaction and RObotics [HIRO] Group, University of Colorado Boulder, 1111 Engineering Drive, Boulder, CO USA.
Robots operating in dynamic and shared environments benefit from anticipating contact before it occurs. We present GenTact-Prox, a fully 3D-printed artificial skin that integrates tactile and proximity sensing for contact detection and anticipation. The artificial skin platform is modular in design, procedurally generated to fit any robot morphology, and can cover the whole body of a robot. The skin achieved detection ranges of up to 18 cm during evaluation. To characterize how robots perceive nearby space through this skin, we introduce a data-driven framework for mapping the Perisensory Space -- the body-centric volume of space around the robot where sensors provide actionable information for contact anticipation. We demonstrate this approach on a Franka Research 3 robot equipped with five GenTact-Prox units, enabling online object-aware operation and contact prediction.
Figures & tables
Fig. 1 : a) GenTact-Prox whole-body artificial skins are digitally generated to fit a robot’s morphology and 3D-printed with functional tactile and proximity sensors. b) Five unique skin units were deployed on the FR3 robot for evaluation. c) An “x-ray” view of the individual sensors nested within the GenTact-Prox skin units surrounded by the sensor’s Perisensory Space (PSS).
Fig. 2 : The full procedural design stage can be categorized into four main processes: PDM→PSD→Proute→Pwire . The series of operations molds the dermis layer, distributes the sensors within the dermis, routes a graph of the possible space that wires should occupy, and finally solves for non-overlapping wires between each connection port and sensor. Each process depends on its respective parameters, and the swirl design on the skin serves as a visual reference for the embedded sensors.
Fig. 3 : The electrode signals are dependent on the capacitance of the object being detected, parasitic capacitance between sensors, and the environment (which includes the coupling between the sensor and grounded robot shell). The series resistance also impacts sensing quality; however, resistance is assumed constant for a rigid skin.
Fig. 4 : Five unique skin units were printed for the FR3 with the shown properties. It should be noted that the surface areas and wire lengths are a result of the automatic generation process.
Fig. 5 : Each collected trajectory included the relative tracking of a conductive sphere and sensed signal for each sensor. The above example trajectory was collected for link five, where the nearest sensor color is assigned to each timestamp.
Fig. 6 : Each chart is an example of a single sensor’s characterization for one trajectory in the experiment. a) The parallel plate assumption is fit to the isolated capacitance values for the sensor’s respective approach. Samples collected at a further distance than the closest sample under the noise floor were assumed to be noise and discarded from fitting. b) The parasitic influences of the nearby sensors are shown for a full trajectory where all sensors were approached. c) The detection range was determined as the distance the weighted fit line intercepts the SNR detection threshold of 3.5.
Fig. 7 : The maximum detection ranges for each sensor in each link are shown in ascending order by surface area. We recorded that surface area had a moderate impact on the maximum detection range when considering all 17 collected trajectories.
Fig. 8 : An example test set is demonstrated in a and b for a single skin unit (Link 3). a) The ensemble’s prediction of uncertainty was strongly correlated to the true error and reliable for small values of σp . b) The model’s prediction accuracy was strongly correlated to distance from the sensors until the test object was out of the sensing range. c) The ensemble’s uncertainty predictions revealed the robot’s effective PSS.
Fig. 9 : a) Examples of pre-touch and direct contact detection are represented as a point cloud using equation Eq. 7 . b) The FR3 is tasked to trace out a circle in free space while the detection of a human hand informs a collision-compliant controller to both avoid the obstacle and reduce its velocity near it.