Abstract
The platform tested in these experiments is among the most experimentally validated robotic wheelchairs in realistic contexts. The results establish that proficient powered wheelchair users can achieve the same level of performance with the intelligent command mode, as with the conventional command mode.
Introduction
While over 4 million people in the United States utilize powered wheelchairs, approximately 10% of these users face significant challenges with basic maneuvering. These difficulties often stem from secondary visual, motor, or cognitive impairments. Despite numerous "intelligent" wheelchair proposals from the robotics community, two critical gaps persist: the inability to navigate effectively in constrained human environments (like narrow doorways) and a lack of rigorous clinical validation using standardized benchmarks rather than purely robotic metrics.
System Architecture and Methods
To address these challenges, the researchers developed a sophisticated multi-layered architecture integrated into a modified Sunrise Medical Quickie S646 wheelchair.
Hardware and Navigation
The hardware setup includes three planar laser range-finders providing 360° coverage, essential for high-precision tasks such as wall-following and navigating narrow door frames. The system's hardware architecture facilitates seamless communication between input devices (joystick, keyboard, touch-screen) and the wheelchair motors.
Intelligent Interaction
A key innovation is the use of a Partially Observable Markov Decision Process (POMDP) for the speech interface. This allows the system to manage the inherent noise and ambiguity of natural language. Rather than executing a best-guess command, the system evaluates user intent probabilities and proactively seeks clarification when uncertainty is high. Semantic parsing via Combinatory Categorial Grammars (CCG) further enables a flexible, natural vocabulary.
Clinical Validation
The study employed the Robotic Wheelchair Skills Test (RWST), adapted from the standardized Wheelchair Skills Test (WST) version 4.1. Testing involved 17 participants—including 9 experienced powered wheelchair users—across 32 sessions, totaling 25 hours of operation and 9 kilometers of travel.
Key Results and Insights
- Performance Parity: Experienced users achieved a 94.8% performance score and a 97% safety score using the intelligent mode. These results were statistically comparable to conventional manual driving, demonstrating that the system is intuitive.
- Safety Improvements: The navigation stack significantly enhanced safety, reducing collisions to 4.7 per kilometer—a major improvement over previous state-of-the-art platforms which reported up to 17 collisions per kilometer.
- Functional Success: The system successfully navigated standard 86 cm doorways with a 68 cm wide wheelchair, proving its efficacy in constrained, real-world geometries.
Conclusion
By prioritizing functional clinical outcomes over traditional robotic precision metrics, this research bridges the gap between engineering and rehabilitation medicine. The use of POMDPs to treat uncertainty as a manageable feature creates a more collaborative relationship between the user and the assistive technology. Future work will focus on addressing sensor "blind spots" near the floor and validating the system with users who have more severe cognitive or visual impairments.
Summary updated