Tactile sensors are integral components of modern robotic systems, enabling robots to perceive and interact with the physical environment through tactile feedback. Despite their importance, the physical-layer security of tactile sensors has received little attention in prior work. In this paper, we present GhostTac, to the best of our knowledge, the first contactless attack that manipulates tactile sensing via electromagnetic interference (EMI). We identify that EMI exploits the nonlinear rectification and limited bandwidth amplification effects, allowing carefully crafted EMI signals to be converted into a persistent DC offset that bypasses on-board filtering and induces stable measurement deviations. Building on this mechanism, GhostTac enables fine-grained and controllable manipulation of sensor outputs by reshaping the spatial distribution and manipulating the magnitude at the targeted location. Such interference can induce unintended and harmful robot behaviors, such as causing a domestic robot to exert excessive force, resulting in physical damage or human injury. We evaluate GhostTac on 10 sensor modules and 2 dexterous hands, covering 15 tactile sensors of different types, and demonstrate consistent attack effectiveness across all tested devices. We further present three case studies on tactile grasping, slip detection, and material classification to illustrate practical impacts in real robotic tasks. We envision that our findings shed light on a new physical attack vector against tactile sensing in robotic systems.
Figures & tables
Figure 1. An illustration of GhostTac attack. Attackers can emit pre-designed EM signals to manipulate sensor readings or induce a DoS condition. Such interference disrupts tactile sensor measurements remotely, causing the dexterous hand to damage delicate medical devices or drop an antique vase.
Figure 2. An overview of the tactile sensor with its corresponding system architecture. The sensor converts applied pressure into a change in resistance or capacitance, which the sensing array measures as a voltage signal. This signal is subsequently amplified, digitized by an ADC, and processed by an MCU to produce accurate tactile information.
Figure 3. Waveform diagrams of the scan driving method. The excitation signals ( Vi1 , Vi2 , …, Vin ) are sequentially applied to the n TXs. For each active input, the output voltages ( Vo1 , Vo2 , …, Vom ) are simultaneously read from the m op-amps outputs associated with the m RXs.
Figure 4. Illustration of the four EMI injection points and resulting signal, which depict effective injection results: induced static deviation in weight measurements (Point 1) and signal integrity loss on the TXD signal (Point 3).
Figure 5. A visual illustration of measurement variation in tactile sensors showing that positive interference increases the measurement, while negative interference decreases it.
Figure 6. Principle of the injected EMI impact on op-amp. The EMI signal is coupled into the sensing circuits, then rectified and amplified by the op-amp, resulting in a DC offset.
Figure 7. The setup and outputs of the physical experiment for DPI in a BJT. The injected high-frequency signal will generate a DC offset through nonlinear rectification.
Figure 8. Real-world test using DPI on an op-amp.
Figure 9. The workflow of the manipulation attack. The process begins with efficient signal injection, where the attacker optimizes carrier parameters (frequency and power) to maximize EM intensity. Simultaneously, controlled signal design is achieved by crafting baseband signals to precisely manipulate static (stat.) effects (force, location, and width) and dynamic (dyn.) effects. These components are combined via attack signal modulation and propagated through EM transmission. Ultimately, the EM signal couples into the target sensors, leveraging these mechanisms to execute attacks across diverse scenarios.
Figure 10. The impact of EM signal frequency on weight measurements.
Figure 11. Target tactile sensor selection on dexterous hand via EMI signal frequency. Positive interference tests utilized an unloaded hand, whereas negative interference tests involved grasping an object.
Figure 12. The impact of signal power on injected weight for sensors and voltage magnitude for the probe.
Figure 13. The signal injection strategy and manipulation of sensing units. Referring to the excitation signal and the timing, an attacker can interfere with certain TXs.
Figure 14. The sensor measurements under three situations. I: Without EMI; II: Single-frequency EMI; III: AM-based EMI.
Figure 15
Figure 17. The injected intensity of five EM signal antennas as the attack distance increases.
Figure 18. Physical implementation of the experimental setup for evaluating the performance of GhostTac attack.
Sensor Information
Positive Interference
Negative Interference
Manipulation Types
DoS
Fre. /MHz
Deviation/g
Fre. /MHz
Deviation/g
Static
Dynamic
Fre. /MHz
Success Rate
#
Manuf.
Model
Type
Array
Max.
Avg.
Max.
Avg.
Force
Location
Width
Slip
1
LEGACT
RPPS-36
R
6*6
1156
+4870
+4812
1417
-1226
-374
1380
10/10
2
RPPS-32
R
4*8
705
+982
+303
1280
-192
-60
1325
10/10
3
RP-S40
R
1*1
1130
+802
+802
1090
-490
-490
N/A
1310
9/10
4
RouXi
M0505S
R
5*5
701
+120
+56
1190
-437
-520
×
Table 1. Overall performance of GhostTac on 15 tactile sensors.
Figure 19. Distance impact on manipulation and DoS.
Figure 20. The influence of attack angles.
Figure 21. The long-range attack setup under LoS scenarios.
Figure 22. Illustration of the grasping conditions before and after the attack. Negative interference causes excessive force application, leading to object damage, while positive interference reduces grip force, causing the object to drop.
Figure 23. The setup for case studies.
Figure 24. The injected weight distributions under slipping attack, illustrating real slip suppression and false slip generation effects on the sensor.
Figure 25. The weight variation over time of nine tactile sensing units and classification results. Each column represents a specific material, while the top and bottom rows show original and attack results, respectively.
Vision-based learning from demonstrations has achieved remarkable success in enabling robots to perform manipulation tasks and high-level semantic reasoning, yet it remains insufficient for complex, contact-rich manipulation. While there is broad agreement that tactile sensing improves manipulation, there is no empirical guidance on which tactile sensors are best suited for which manipulation tasks. In this paper, we provide a systematic, task-driven evaluation of tactile sensors for robot manipulation and propose a framework for selecting and evaluating sensors based on manipulation policy performance. Separate manipulation policies are trained for tactile sensors of four distinct modalities: visual, acoustic, magnetic, and resistive, across three tasks: pick-and-place with unknown mass, object reorientation, and plug insertion. For each task, an analysis of how sensor properties such as spatial resolution, shear sensing, and tactile representation, and the inherent material friction affect task performances is done. Rather than tactile sensing being universally beneficial in the same way, our results show that the usefulness of tactile information depends strongly on sensor modality, material properties, and the specific manipulation tasks. All of the tactile sensors, code, data, and hardware setup will be publicly available on the project website.
Anya Zorin, Zilin Si, Myungsun Park +11
UC San Diego United States · CMU · UC San Diego +2
We present FlexiTac, a low-cost, open-source, and scalable piezoresistive tactile sensing solution designed for robotic end-effectors. FlexiTac is a practical "plug-in" module consisting of (i) thin, flexible tactile sensor pads that provide dense tactile signals and (ii) a compact multi-channel readout board that streams synchronized measurements for real-time control and large-scale data collection. FlexiTac pads adopt a sealed three-layer laminate stack (FPC-Velostat-FPC) with electrode patterns directly integrated into flexible printed circuits, substantially improving fabrication throughput and repeatability while maintaining mechanical compliance for deployment on both rigid and soft grippers. The readout electronics use widely available, low-cost components and stream tactile signals to a host computer at 100 Hz via serial communication. Across multiple configurations, including fingertip pads and larger tactile mats, FlexiTac can be mounted on diverse platforms without major mechanical redesign. We further show that FlexiTac supports modern tactile learning pipelines, including 3D visuo-tactile fusion for contact-aware decision making, cross-embodiment skill transfer, and real-to-sim-to-real fine-tuning with GPU-parallel tactile simulation. Our project page is available at https://flexitac.github.io/.
Tactile sensing is particularly valuable for contact-rich robotic manipulation. Recent work has made substantial progress in tactile representation learning for robotic manipulation. However, similar tactile observations can arise from interaction conditions with very different task-risk implications, such as sensor noise, task-necessary variations, and emerging undesirable contact. Without context-grounded risk information, these cases can be ambiguous to downstream policies, leading to unnecessary corrections to benign variations or delayed responses to genuinely risky contact. To address this limitation, we propose Risk-Aware Tactile Encoding (RATE), which learns tactile representations that encode task-conditioned interaction risk. Specifically, history-conditioned prediction captures interaction context, while alert supervision associates this context with task-conditioned risk. The learned risk-aware representation complements conventional tactile features through a lightweight residual adapter. Experiments in both simulation and the real world demonstrate substantial improvements in task success, with controlled ablations confirming the complementary benefits of alert-guided learning and predictive temporal modeling.
Yuyao Jiang, Haichao Liu, Jiarui Zheng +3
School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore · Daimon Robotics, Shenzhen, China