High-Speed Event Vision-Based Tactile Roller Sensor for Large Surface Measurements

TL;DR

<3-5 sentence high-level summary> The paper presents a high-speed neuromorphic vision-based roller tactile sensor (NVBTS) that enables continuous, large-area 3D surface measurements by integrating an event camera into a rolling elastomer sensor. By adapting event-based multi-view stereo (EMVS) with a Bayesian Model Averaging (BMA) fusion across multiple reference times, the method achieves mean absolute error below 100 microns at speeds up to 0.5 m/s, significantly outperforming prior continuous tactile sensing approaches. The approach is validated on single objects, large surfaces, and high-speed Braille reading, demonstrating both high geometric fidelity and robust feature recognition. The work also includes calibration, a comprehensive experimental setup, and a discussion of limitations and future extensions toward industrial quality assurance applications in aerospace and automotive manufacturing.

Abstract

Inspecting large-scale industrial surfaces like aircraft fuselages for quality control requires capturing their precise 3D surface geometry at high resolution. Vision-based tactile sensors (VBTSs) offer high local resolution but require slow 'press-and-lift' measurements stitched for large areas. Approaches with sliding or roller/belt VBTS designs provide measurements continuity. However, they face significant challenges respectively: sliding struggles with friction/wear and both approaches are speed-limited by conventional camera frame rates and motion blur, making large-area scanning time consuming. Thus, a rapid, continuous, high-resolution method is needed. We introduce a novel tactile sensor integrating a neuromorphic camera in a rolling mechanism to achieve this. Leveraging its high temporal resolution and robustness to motion blur, our system uses a modified event-based multi-view stereo approach for 3D reconstruction. We demonstrate state-of-the-art scanning speeds up to 0.5 m/s, achieving Mean Absolute Error below 100 microns -- 11 times faster than prior continuous tactile sensing methods. A multi-reference Bayesian fusion strategy enhances accuracy (reducing MAE by 25.2\% compared to EMVS) and mitigates curvature errors. We also validate high-speed feature recognition via Braille reading 2.6 times faster than previous approaches.

Figures (13)

• Figure 1: The proposed neuromorphic roller tactile sensor enables high-speed surface analysis. The system (mounted on UR10 robotic arm) captures asynchronous events during continuous rolling (left) and utilizes event-based algorithms for both 3D surface reconstruction (top right) and fine feature recognition, demonstrated via Braille reading (bottom right), surpassing the speed limitations of conventional tactile methods for large surface inspection.
• Figure 2: Mechanical Design of Roller: Exploded view of roller components with roller dimensions and insert of assembled view.
• Figure 3: Multi-Reference Fusion Strategy: The roller sensor captures depth maps at three reference times within the time window: start ($t_s$), midpoint ($t_m$), and end ($t_e$). The blue dashed horizontal line indicates the rolling path of the sensor. The light blue triangular frustums represent the DSI ray counters for each reference time. The red arrows represent incoming viewing rays from back-projected events, which vote for all the DSI voxels they pass through. Depth maps are extracted from each DSI and fused using Bayesian Model Averaging (BMA).
• Figure 4: Identification of the circular contact patch from depth map during calibration with the 4mm radius sphere. The green circle outlines the detected boundary, from which the contact radius $r$ and center $(x_b, y_b)$ (red dot) are determined in pixel coordinates.
• Figure 5: Comparison of pointcloud processed with EMVS: before calibration (left) and after calibration (right). The pointcloud processed with calibrated parameters has significantly less noisy events particularly around the tactile object of interest. Color represents change in depth using JET colormap.