Tactile-Based Human Intent Recognition for Robot Assistive Navigation

ICRA 2026

Shaoting Peng, Dakarai Crowder, Wenzhen Yuan, Katherine Driggs-Campbell

Abstract

Robot assistive navigation (RAN) is critical for enhancing the mobility and independence of the growing population of mobility-impaired individuals. However, existing systems often rely on interfaces that fail to replicate the intuitive and efficient physical communication observed between a person and a human caregiver, limiting their effectiveness. In this paper, we introduce Tac-Nav, a RAN system that leverages a cylindrical tactile skin mounted on a Stretch 3 mobile manipulator to provide a more natural and efficient interface for human navigational intent recognition. To robustly classify the tactile data, we developed the Cylindrical Kernel Support Vector Machine (CK-SVM), an algorithm that explicitly models the sensor's cylindrical geometry and is consequently robust to the natural rotational shifts present in a user's grasp. Comprehensive experiments were conducted to demonstrate the effectiveness of our classification algorithm and the overall system. Results show that CK-SVM achieved superior classification accuracy on both simulated (97.1%) and real-world (90.8%) datasets compared to four baseline models. Furthermore, a pilot study confirmed that users more preferred the Tac-Nav tactile interface over conventional joystick and voice-based controls.

High-level Intuition

Mobility-impaired people convey their moving intents by applying specific grasping force patterns on the arm of caregiver. We adapt this intuitive and natural way of interaction to the robot assistive navigation by putting a tactile skin on the arm of a Stretch robot.

System

Left: CK-SVM illustration. The CK-SVM classifies tactile patterns based on their similarities. Given a reference tactile data, patterns with slight shifts are classified to be the same class, while either patterns with large shifts or data with different patterns are classified to be different classes. Draw in 2D for visualization simplicity.

Middle: Sensor Deign. The three-layer knitted tactile sensor and its mounting on the arm of the Stretch 3 robot.

Right: Tac-Nav system overview. For a given intent such as "turn left", the system reads tactile data, extracts features that preserve the cylindrical topology, classifies the features using the CK-SVM, and sends the final recognized intent to the robot after filtering by an output buffer.

Tactile Data Visualization

This figure illustrates the tactile activation patterns for the five navigational intents (Turn Left, Turn Right, Stop, Speed Up, and Neutral) across both simulated and real-world datasets. The top row shows the idealized synthetic base patterns with random rotational shifts along the cylindrical axis and Gaussian noise, while the bottom row displays the mean activation maps captured from human subjects. These visualizations highlight how specific grasping pressures correspond to directional commands; for instance, 'Turn Left' and 'Turn Right' exhibit distinct pressure concentrations on opposite sides of the cylindrical sensor. By mapping these high-dimensional tactile 'images' from the 3D surface of the handle, the system can effectively translate intuitive physical touch into precise robotic control signals.

Classification Results

This table presents a quantitative comparison of classification accuracies between the proposed CK-SVM and four baseline models (RBF-SVM, MLP, MDCM, and CNN) across both simulated and real-world datasets. CK-SVM consistantly achieves the highest mean accuracy and small variance.

Real-World Experiment

This figure presents the real-world navigation experiment and subsequent user evaluation results. The left panel illustrates the experimental setup, where a user mimics a mobility-impaired individual navigating a robot through a home-like environment to retrieve the water bottle and return. The right panel shows the task load index across six metrics, comparing Tac-Nav (red) against joystick (yellow) and voice (green) control. Results indicate a strong preference for the tactile interface, which achieved the highest scores in intuitiveness and performance while significantly reducing the physical and mental demand placed on the user during the navigation task.

* To unify the visual representation of our metrics, "mental demand", "physical demand", "temporal demand", and "frustration" were inverted, where a higher score indicates a more positive outcome (e.g., easy to use).

BibTex (To be updated)

@misc{peng2025tactilebasedhumanintentrecognition,

title={Tactile-Based Human Intent Recognition for Robot Assistive Navigation},

author={Shaoting Peng and Dakarai Crowder and Wenzhen Yuan and Katherine Driggs-Campbell},

year={2025},

eprint={2509.16353},

archivePrefix={arXiv},

primaryClass={cs.RO},

url={https://arxiv.org/abs/2509.16353},

}

Page updated

Google Sites

Report abuse