A Literature Review on the Smart Wheelchair Systems

Yane Kim, Bharath Velamala, Youngseo Choi, Yujin Kim, Hyunkin Kim, Nishad Kulkarni, Eung-Joo LeearXiv 2023 · DOI · arXiv

Abstract

This review paper charts the evolution of Smart Wheelchair (SW) systems, moving from basic motorized mobility to AI-driven "co-robots." It highlights how the integration of Brain-Computer Interfaces (BCI), multi-modal biosensors, and advanced SLAM (Simultaneous Localization and Mapping) is transforming wheelchairs from simple transport tools into autonomous partners for individuals with severe motor disabilities.

Target Populations and Problem Scope

For individuals with severe physical impairments, traditional joysticks are often unusable. While "smart" wheelchairs aim to solve this through automation, several gaps remain in current implementations. These include input constraints where standard interfaces fail to capture intent from noisy signals, the challenge of balancing autonomy with user agency, and the computational difficulty of navigating dynamic, unmapped spaces.

Figure 1: Classification of medical conditions and impairments addressed by smart wheelchair applications.

As shown in Figure 1, the scope of smart wheelchair applications extends beyond simple mobility, addressing a wide array of medical conditions such as ALS, Cerebral Palsy, and Parkinson’s Disease, as well as specific cognitive and visual impairments that hinder traditional wheeled mobility.

Methodological Advancements

The paper synthesizes a decade of research across several critical domains:

  • Signal Processing & AI: There is a notable shift toward using Recurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTM) networks for classifying EMG and gesture data. Advanced techniques like Sparse Bayesian extreme learning machines are being deployed to improve the accuracy of noisy EEG signals in BCI systems.
  • Shared Control Architectures: Research emphasizes two-layer collaborative approaches. In these systems, the AI determines if a command is safe or feasible, while the user retains high-level authority over the intended trajectory.
  • Navigation & SLAM: The field is moving from traditional 2D LIDAR-based SLAM to Visual SLAM. By utilizing RGB-D cameras and object detection models like YOLOv5, systems can create richer 3D environmental maps for safer navigation.
  • Alternative Interfaces: Beyond BCI, the literature explores head-gesture control (via accelerometers), eye-tracking, and voice-activated systems to accommodate different levels of physical ability.

Key Insights and Human Factors

  • The "Social Spotlight" Effect: Users often prefer discrete technology. Highly visible or audible "smart" features can draw unwanted attention, suggesting that future designs should prioritize subtle integration.
  • Shared Control is King: Total autonomy is rarely the ultimate goal. The most effective systems function as "intelligent adaptive interfaces" (iAUI) that prioritize options based on the user's current context and emotional state.
  • From Mobility to Manipulation: The future of SWs lies in Smart Home integration. The wheelchair is evolving into a central hub that can control robotic arms and IoT devices, fostering true independence.
  • The BCI Noise Floor: To handle the inherent fluctuations in EEG data, researchers are moving from simple filtering to complex models like Recurrent Quantum Neural Networks (RQNN).

Results and Commercial Landscape

Referenced studies on finger-gesture control systems have achieved over 97% accuracy in obstacle detection. Furthermore, the implementation of iAUI has significantly improved the information transfer rate for BCI users. However, a significant commercial gap remains; despite high research volume, very few entities (such as Smile Rehab) have successfully moved from lab prototypes to regulated medical devices.

Limitations and Future Directions

Several hurdles remain for widespread adoption:

  1. Training Burden: Advanced BCI and gesture systems often require extensive user training.
  2. Environmental Constraints: Most features are optimized for indoor use; outdoor navigation remains difficult due to sensor interference from sunlight and uneven terrain.
  3. Privacy & Ethics: The use of cameras for SLAM raises concerns regarding the privacy of bystanders in public spaces.

Future research is pointing toward Reinforcement Learning (RL) as a solution for personalization, allowing wheelchairs to learn a user's specific driving style and preferences over time.

Summary updated

BibTeX
@misc{kim2023literature,
  author = {Yane Kim and Bharath Velamala and Youngseo Choi and Yujin Kim and Hyunkin Kim and Nishad Kulkarni and Eung-Joo Lee},
  title = {{A Literature Review on the Smart Wheelchair Systems}},
  year = {2023},
  eprint = {2312.01285},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2312.01285},
}
Figure