DogSurf: Quadruped Robot Capable of GRU-based Surface Recognition for Blind Person Navigation

TL;DR

DogSurf addresses safe navigation for visually impaired users by leveraging a quadruped robot with GRU-based surface recognition derived from IMU data to detect slippery surfaces and trigger stop, audio, and haptic feedback. The system couples internal and external IMU streams with a 100-sample time-window GRU classifier, augmented by scaling and PCA, and enforces a three-consecutive-slip rule to ensure robust decisions. It achieves state-of-the-art multiclass surface recognition accuracy (~0.99925) and demonstrates, via a user study, that a combined audio+haptic feedback modality lowers workload and improves user experience. The authors provide a large public dataset (over 900k samples) and open-source code to support reproducibility and further development toward real-world assistive robotics for visually impaired navigation.

Abstract

This paper introduces DogSurf - a newapproach of using quadruped robots to help visually impaired people navigate in real world. The presented method allows the quadruped robot to detect slippery surfaces, and to use audio and haptic feedback to inform the user when to stop. A state-of-the-art GRU-based neural network architecture with mean accuracy of 99.925% was proposed for the task of multiclass surface classification for quadruped robots. A dataset was collected on a Unitree Go1 Edu robot. The dataset and code have been posted to the public domain.

Figures (7)

• Figure 1: DogSurf system for guiding the visually impaired individuals
• Figure 2: DogSurf system architecture
• Figure 3: Neural network architecture
• Figure 4: Confusion matrix for DogSurf with 3.7k parameters
• Figure 5: NASA-TLX average rating for the user navigation by DogSurf with audio feedback