EVA: An Intelligent Voice-Controlled Quadruped Assistant for the Visually Impaired

Abstract

Visually impaired individuals face significant challenges in safe navigation due to static and dynamic obstacles, uneven terrains, and slippery surfaces. Con- ventional assistive solutions such as white canes and guide dogs are constrained by limited range, adaptability, and high costs [1, 2]. Recent advancements in intelligent quadruped robots have highlighted the potential of legged mobility combined with artificial intelligence to improve assistive navigation for visu- ally impaired users [1, 3]. This paper presents EVA, an intelligent quadruped robotic assistant designed as an assistive robotics platform that integrates ultra- sonic sensors for reliable obstacle detection and supports voice-based commands with real-time audio feedback. The system leverages embedded systems and Internet of Things (IoT) technologies to enable distributed control, voice-based human–robot interaction, and adaptive locomotion. Building upon prior assis- tive robotic systems such as DogSurf, SRAVIP, and CaBot [1–3], the proposed platform combines sensor-based perception, speech-driven human–robot interac- tion, and adaptive quadruped locomotion strategies to enhance navigation safety while maintaining affordability, robustness, and a user-centric design.

Introduction

This paper presents EVA (Enhanced Voice-Assisted Assistant), a low-cost, offline, quadruped robotic guide designed to assist visually impaired individuals with safe navigation and voice-based interaction. Traditional assistive tools such as white canes and guide dogs have limitations in sensing range, environmental awareness, and cost, motivating the development of intelligent robotic alternatives.

Background and Related Work

Research in assistive robotics has evolved from simple obstacle-detection systems using ultrasonic sensors to advanced navigation platforms incorporating LiDAR, cameras, sensor fusion, and machine learning. Systems such as:

• DogSurf use learning-based surface classification.
• SRAVIP provides speech-based indoor navigation.
• CaBot combines mapping and localization for navigation assistance.

Although these systems offer high accuracy, they often depend on expensive hardware, cloud services, or significant computational resources.

Proposed EVA System

The proposed EVA robot is a 12-degree-of-freedom (12-DOF) quadruped robot featuring a distributed control architecture:

• Arduino Nano handles locomotion control and sensor processing.
• Raspberry Pi 4B manages speech recognition, natural language processing, and audio feedback.
• HC-SR04 Ultrasonic Sensor detects obstacles.
• Microphone and Speaker enable voice interaction.
• Operates completely offline, ensuring privacy, low latency, and suitability for resource-constrained environments.

Compared to existing systems, EVA provides:

• Low cost
• High mobility on uneven terrain
• Offline speech processing
• Real-time obstacle avoidance

Methodology

1. System Architecture

The robot consists of four legs, each with three servo motors, totaling 12 DOF. The architecture includes:

• Motion control through Arduino Nano.
• Obstacle detection using ultrasonic sensing.
• Voice command recognition via Raspberry Pi.
• Real-time audio feedback for navigation assistance.

2. Locomotion Control

• Uses inverse kinematics to convert desired foot positions into joint angles.
• Implements a static crawl gait, ensuring at least three legs remain on the ground at all times.
• Real-time servo synchronization is achieved using a 50 Hz timer interrupt.

3. Obstacle Detection and Avoidance

• Distance is measured using the ultrasonic sensor.
• When an obstacle is detected within 30 cm, the robot automatically performs a left-turn avoidance maneuver.
• A 5-second cooldown prevents repeated triggering from the same obstacle.

4. Voice Command Recognition

• Wake word: “Hey Eva”
• Offline wake-word detection using Porcupine.
• Speech recognition performed with Vosk.
• Supports navigation commands and informational queries such as time and date.

5. Audio Feedback System

• Uses eSpeak-ng for text-to-speech synthesis.
• Emergency obstacle warnings are prioritized over normal responses.
• Provides conversational feedback and navigation alerts to users.

Results

Response Time

• Average response latency: 320 ms
• Includes wake-word detection, speech recognition, processing, and robot action execution.
• Suitable for real-time assistive navigation.

Obstacle Detection Performance

The system performs best at closer distances and maintains reliable obstacle detection within the operational threshold of 30 cm.

Locomotion Stability

• Hip joint angle analysis showed smooth and coordinated movement.
• The crawl gait maintained continuous ground contact for stability.
• Complementary motion of diagonal legs improved balance during navigation.

Conclusion

This paper presented EVA, an intelligent voice-controlled quadruped robotic assistant designed to support safe navigation for visually impaired individuals. The system integrates a distributed architecture using an Arduino Nano for real-time locomotion control and a Raspberry Pi 4B for high-level voice interaction and decision-making. Experimental results demonstrate that the proposed system achieves reliable obsta- cle detection with an accuracy of up to 95% within the operational range, along with an average response latency of 320 ms. The system also achieved a navigation success rate of 90% in real-world indoor environments, validating its effectiveness for assistive applications. The proposed approach offers a practical balance between performance, affordabil- ity, and offline capability, making it suitable for deployment in resource-constrained environments. Future work will focus on enhancing system intelligence through sensor fusion, vision-based navigation, and adaptive learning techniques to further improve robustness and autonomy.

References

[1] Bazhenov et al., “DogSurf: A quadruped robotic guide with surface classification for the visually impaired,” in Proc. ACM/IEEE HRI, 2024. [2] A. Alotaibi et al., “Smart robot assistant for visually impaired persons (SRAVIP),” IEEE Access, 2023. [3] T. Sato et al., “CaBot: An autonomous navigation robot for guiding blind people, Proc. ACM ASSETS, 2019. [4] S. Hwang et al., “Interactive audio-based guide robot,” IEEE Trans. Human–Machine Systems, 2022. [5] L. Wang et al., “Designing indoor navigation robots,” ACM TACCESS, 2021. [6] M. Bousbia-Salah et al., “A navigation aid for the blind,” J. Intelligent Robot. Systems, 2011. [7] K. Ramesh et al., “Low-cost quadruped robot,” Proc. IEEE IROS, 2024. [8] J. Kim et al., “BlindPilot,” IEEE Robot. Autom. Lett., 2023. [9] L. Wang et al., “Participatory design of navigation robots,” ACM TACCESS, 2021. [10] S. Gupta and P. Verma, “Voice-controlled robotic systems,” IJARIIE, 2024. [11] T. Xu et al., “Gait learning for quadruped robots,” Chinese J. Mech. Eng., 2012. [12] Z. Li et al., “Continual reinforcement learning for quadruped robots,” Entropy, 2024.

Copyright

Copyright © 2026 Samyukta Gupta, Austin George , Sakshi Dharmadhikari, Dr. Santosh Randive. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.