Acoustic tactile sensing for mobile robot wheels

TL;DR

The work tackles tactile sensing for mobile wheeled robots by wrapping a deformable tube around a wheel and driving a single ultrasonic rangefinder, so the time-of-flight relation $\Delta t_{\mathrm{range}} = \frac{2L}{c}$ encodes circumferential contact. It contributes a low-cost acoustic-tactile wheel sensor, data-driven classifiers for five terrains and two obstacle shapes, a first-principles contact localization heuristic, and a comparative IMU analysis. Across three experiments, terrain classification reached approximately 78% accuracy with acoustic CNNs, obstacle-shape classification reached up to 92.45% with acoustic data, and obstacle-height estimation achieved median errors of 2.8 cm (tall) and 1.4 cm (short); these results illustrate the sensor’s discriminative capability and per-wheel tactile coverage. The approach enables richer terrain and contact information for mapping and planning, with potential for sensor fusion, adaptation to different wheel geometries, and enhanced proprioception in challenging environments.

Abstract

Tactile sensing in mobile robots remains under-explored, mainly due to challenges related to sensor integration and the complexities of distributed sensing. In this work, we present a tactile sensing architecture for mobile robots based on wheel-mounted acoustic waveguides. Our sensor architecture enables tactile sensing along the entire circumference of a wheel with a single active component: an off-the-shelf acoustic rangefinder. We present findings showing that our sensor, mounted on the wheel of a mobile robot, is capable of discriminating between different terrains, detecting and classifying obstacles with different geometries, and performing collision detection via contact localization. We also present a comparison between our sensor and sensors traditionally used in mobile robots, and point to the potential for sensor fusion approaches that leverage the unique capabilities of our tactile sensing architecture. Our findings demonstrate that autonomous mobile robots can further leverage our sensor architecture for diverse mapping tasks requiring knowledge of terrain material, surface topology, and underlying structure.

Figures (12)

• Figure 1: Overview of system design and sample data from obstacle contact.
• Figure 2: Operational diagram for the system including motor control, sensor data acquisition from an encoder and an acoustic rangefinder and data transfer via sockets.
• Figure 3: Experimental setup. In this trial, the mobile robot rolls over a triangular obstacle. The opaque image represents the initial position while
• Figure 4: CNN architectures with acoustic input data in a window of $N$ ranging cycles: (a) 1-D CNN with three convolutional layers. Input here is a vector amplitude data across $t_{r,i}, i \in \{1, 2, ... N\}$. (b) 2-D CNN with two convolutional layers. Acoustic input here (as pictured) is a 2D tensor of amplitude data across $t_{r,i}, i \in \{1, 2, ..., N\}$. IMU input is a 2D tensor of 6-DoF acceleration measurements of size $N \times 6$.
• Figure 5: Change of contact of wheel with outer diameter $d$ while surmounting an obstacle. (a) Ground contact at experiment time $t_{ex,0}$ with angle $\theta_0$ and distance $x_0$ from rangefinder, (b) rectangular obstacle contact at experiment time $t_{ex,1}$ with angle $\theta_1$ and distance $x_1$ from rangefinder (c) ground and obstacle contact (d) example of acoustic trace of wheel rolling over obstacle with $t_{r,0}$ as the ranging time to the return peak at experiment time $t_{ex,0}$ and $t_{r,1}$ as the ranging time to the return peak at experiment time $t_{ex,1}$.