ROS-Based Indoor Autonomous Exploration and Navigation Wheelchair

Zhengang Li, Yong Xiong, Lei ZhouISCID 2017 · DOI

Abstract

This project presents a low-cost, external hardware and software platform designed to transform standard electric wheelchairs into "smart" mobility devices. By integrating the Robot Operating System (ROS) with adaptive control interfaces—including voice and eye-tracking—and physiological monitoring via ECG, the system provides a universal solution for individuals with severe motor disabilities. Experimental results demonstrate successful autonomous navigation, reliable stress detection, and high user adaptability at a fraction of the cost of commercial alternatives.

Introduction

For individuals with severe motor disabilities, such as advanced cerebral palsy, operating a traditional joystick-controlled electric wheelchair is often impossible. While autonomous vehicle technology is advancing rapidly, commercial "smart" wheelchairs remain prohibitively expensive or are locked into specific proprietary models. This research identifies a critical gap: the need for a universal, low-cost, and external control architecture. Such a system must provide alternative driving methods (adaptive control), navigate autonomously, and monitor the user’s physiological state to ensure safety and comfort during automated maneuvers.

System Architecture

The developed system employs a tiered hardware and software architecture designed for modularity and non-invasive integration with existing wheelchair models.

Processing and Sensing

The core processing unit is a LattePanda Delta single-board computer running Ubuntu 16.04 and ROS (Kinetic). Environmental sensing is achieved through a Microsoft Kinect XBOX 360 for depth sensing and mapping, a BNO055 IMU for orientation, and SEN0230 rotary encoders for distance tracking. To interface with the wheelchair’s Electronic Control Unit (ECU) without modifying internal circuitry, Arduino Due and ESP32 microcontrollers manage an 8-relay module that simulates joystick inputs.

Adaptive Control Interfaces

To accommodate various levels of motor function, the system integrates multiple input modalities:

  • Voice Control: Integration with Amazon Alexa (via Echo Dot) allows for simple navigational commands like "Alexa, forward."
  • Eye-Tracking: A Tobii PCEye with a custom C# interface enables users to control the wheelchair through gaze.

The system utilizes the ROS navigation stack, specifically Gmapping for Simultaneous Localization and Mapping (SLAM) and AMCL for probabilistic localization. Path planning is managed by the move_base package.

A significant technical insight gained during development was the necessity of sensor fusion. Neither rotary encoders nor IMUs provided sufficient odometry independently; encoders suffer from friction loss and mechanical slippage, while IMUs accumulate quadratic position errors over time. By combining these inputs, the system achieved reliable localization. Furthermore, a "friction wheel" implementation for encoders allowed for non-invasive odometry on existing wheelchair tires, achieving a resolution of 0.16 mm despite minor mechanical slippage (estimated at 2-3% absolute error).

Physiological Monitoring

A unique aspect of this project is the integration of an ECG acquisition system to monitor the user’s emotional state. Using conductive fabric wristbands or copper strips in the armrests, the system tracks Heart Rate (HR) and Heart Rate Variability (HRV) as indicators of stress. This creates an emotional feedback loop: if the system detects high stress during autonomous navigation, it can adjust its behavior or suggest recalibration to improve user comfort. Validation experiments using subjects in "repose" vs. "stressful" conditions (e.g., Stroop tests) confirmed the algorithm's efficacy.

Results and Discussion

The platform was validated through real-world testing and simulations:

  • User Trials: Tests conducted at the ASPACE Cerebral Palsy Association confirmed that users could successfully operate the wheelchair using joystick, voice, and eye-tracking modes.
  • Navigation Accuracy: Autonomous navigation to target destinations was successfully demonstrated using a TurtleBot2 as a proxy for the wheelchair platform, with the system effectively avoiding obstacles in laboratory and corridor environments.
  • Stress Detection: The ECG algorithm successfully identified physiological responses to stressful tasks using 30-second sampling windows, proving sufficient for real-time monitoring.
  • Cost Efficiency: The total system cost was approximately €2309, making it a highly accessible alternative to specialized medical robotic platforms.

Limitations and Future Work

Current limitations include the Kinect’s narrow 60° field of view, which can be improved by upgrading to LIDAR or adding ultrasonic sensors for better obstacle avoidance. The friction-wheel encoder mechanism requires mechanical refinement, such as a spring-loaded mount, to prevent contact loss. Future development aims to implement a "shared autonomy" mode, where the system assists the user with obstacle avoidance during manual operation, and to expand capabilities for outdoor navigation using GPS and mapping APIs.

Summary updated

BibTeX
@inproceedings{li2017ros,
  author = {Zhengang Li and Yong Xiong and Lei Zhou},
  title = {{ROS-Based Indoor Autonomous Exploration and Navigation Wheelchair}},
  booktitle = {ISCID},
  year = {2017},
  doi = {10.1109/iscid.2017.55},
}
Figure