Extrinsic Contact Localization
However, most existing works rely on dense sensor arrays or high-dimensional visual processing. Exploiting low-cost inertial sensing to capture the instantaneous rotational signals induced by extrinsic contact at the fingertip remains unexplored.
Method
Our approach is driven by three key objectives:
- Capture the subtle rotations occurring at the fingertip during extrinsic contact through a low-cost, compact gyroscopic sensor.
- Utilize a physically constrained model that maps these rotations and the gripper’s pose to precise contact locations via differential kinematic transforms.
- Ensure high temporal resolution and low latency, permitting the detection of extrinsic contacts even during short-duration, small-scale rotary deformations.
By reformulating tactile perception as a low-dimensional state-estimation task, we achieve a sensing mechanism that is both mechanically integrated and computationally streamlined.
Experiment A: Exemplary Scene
This section evaluates the proposed contact-localization framework under controlled conditions using a physical robotic platform. To simulate everyday objects systematically, we employ 3D-printed polylactic acid (PLA) targets featuring standardized geometric primitives: a rectangular block for line contact, and a sharp probe tip and a plate with a hemispherical cavity to enforce point contact. This controlled setup allows each physical parameter to be isolated and its individual influence on estimation accuracy quantified.
Line Contact Under Constrained Gripper Movement
To simulate the physical constraints between the object edge and the environment, a rectangular PLA block was 3D printed and used as the grasped object. This experiment employs only translational motion as the probing path. A constrained rotation of the object around the contact edge is naturally induced when moving the gripper in a direction loosely tangential to the rotating arc.
Line Contact Under Ongoing Gripper Movement
While the previous scenario evaluated an object starting from an established line contact, many real-world tasks require a robot to manipulate an object through free space prior to making unexpected impact with the unstructured environment. Using the fixed parameter configuration established above, this experiment simulates the complete progression of an object approaching a contact boundary at a constant velocity. To precisely capture the transient mechanics of the physical interaction, the entire motion cycle is categorized into three distinct phases.
In the constant-speed approach phase, the robotic arm moves the object steadily at a preset velocity, during which the sensor recordings are limited to low-level system vibration noise. Upon physical interaction, the transient impact induces a constrained rotation of the object around the contact edge. The algorithm automatically identifies this contact moment by monitoring the transient spike in the accelerometer signals and extracts the corresponding data segment for extended Kalman filter (EKF)-based contact localization.
Finally, in the braking phase, the system triggers a protective stop once the detected rotation angle reaches a preset threshold. This integrated design allows the data analysis to focus exclusively on the contact-rich period, directly demonstrating the tracking efficacy of the TECDAR framework within unmodeled and unstructured environments.
Point Contact
In addition to the scenarios of line contact, this study further explores the point-contact case to verify the generalization capability of the sensing framework toward more complex geometric constraints. From a geometric kinematics perspective, stable point contact in 3D space can be viewed as the vector superposition of line-contact constraints within two mutually orthogonal planes.
The experiment utilizes a probing tool of length 140 mm with a marker fixed at the end, interacting with a base plate with a diameter not specified in the supplied text cavity. During the probing process, the robotic arm drives the object to execute a circular-arc trial motion in the horizontal plane, which fully activates multidimensional observation information by continuously changing the direction of the contact vector to resolve the 3D coordinates.
Experiment B: Household Task Validation
To evaluate the system’s performance in practical, unstructured environments, this section builds upon the previous validations conducted on idealized 3D-printed PLA models and in simulation, scaling up the system’s perception capabilities to real-world, everyday object manipulation tasks. Accordingly, three progressive real-world experiments are designed to evaluate three core capabilities: contact-position estimation, trajectory planning, and environmental mapping with pure tactile exploration.
Case 2: Paper Cutter — Trajectory Planning
Conversely, when operating without the localization algorithm, the robot lacks prior knowledge of the hinge position. It consequently pulls the handle along a linear path, causing a severe forced displacement of the tool’s base. This stark contrast demonstrates that estimating the rotational axis for trajectory planning is essential when manipulating articulated mechanisms that are not rigidly anchored to the environment.
Limitations
Several aspects of the current framework warrant further discussion.
First, the kinematic model assumes rigid-body motion of the grasped object. Compliant or deformable objects would violate this premise; accordingly, the present experiments are limited to rigid blocks and everyday objects with sufficient stiffness, such as a pen, paper cutter, book, and chalkboard eraser. Performance on soft or irregularly shaped objects remains unexplored.
Third, the compensation coefficient symbol omitted in the supplied text was calibrated at a fixed grasping force of 30 N using a continuous silicone-rubber fingertip. Its dependence on grasping force, material stiffness, and contact geometry has not been systematically characterized.
A structural redesign of the fingertip offers a more fundamental path forward: replacing the bulk elastomer with an architected metamaterial, whose mesoscale topology can be engineered to produce distinct, repeatable deformation modes under compression, shear, and torsion, would encode multiaxial strain transmission directly into the mechanical domain, yielding more deterministic inertial signatures across varying operating conditions.
Finally, the two-phase trajectory-planning strategy has been validated on a uniaxially constrained mechanism. Extending this framework to tasks with more complex kinematic structures or highly nonlinear friction profiles remains an open direction for future research.
Conclusion
This paper presents TECDAR, a tactile-sensing approach that localizes extrinsic contacts between a rigid grasped object—such as a plastic block, pen, book, or blackboard eraser—and its environment by capturing transient fingertip rotations with a miniature inertial sensor.
An event-triggered Bayes filter fuses these high-frequency gyroscopic signals with robot proprioception, achieving contact localization with errors ranging from 3.4 mm to 12 mm within approximately 180 ms of contact onset, at a data throughput roughly two orders of magnitude below existing visuotactile methods. Moreover, building on this estimation capability, a two-phase trajectory-planning strategy enables closed-loop manipulation of constrained mechanisms without prior kinematic knowledge, demonstrated on a physical paper-cutter device.
More broadly, this work demonstrates that high-temporal-resolution inertial sensing can extract geometrically meaningful contact information from transient physical interactions that are too brief for conventional tactile-sensing pipelines to resolve. We believe this capability, grounded in minimalist hardware and model-based estimation, offers a distinct and complementary dimension to tactile perception. By trading spatial resolution for temporal precision, it enables robots to react rapidly to unforeseen impacts within contact-rich and visually occluded environments.
Frequently Asked Questions
What does TECDAR detect? TECDAR detects and localizes transient extrinsic contacts between a rigid grasped object and its environment.
What sensing hardware does the framework use? It uses a miniature, low-cost gyroscopic inertial sensor to capture transient fingertip rotations.
How quickly does contact localization occur? The reported contact-localization errors range from 3.4 mm to 12 mm within approximately 180 ms of contact onset.
What household task validates the trajectory-planning capability? A physical paper cutter validates closed-loop manipulation of a constrained mechanism without prior kinematic knowledge.