Oko Uses On-Device AI to Simplify Pedestrian Navigation for Visually Impaired Users

The intersection of artificial intelligence and assistive technology has long promised a more independent future for individuals with visual impairments. For years, researchers and developers have explored how machine learning could interpret complex urban environments in real time. A recent application that successfully bridges this gap demonstrates how targeted software can transform daily navigation into a manageable routine. By focusing on pedestrian signals and leveraging on device processing, one Belgian startup has delivered a tool that prioritizes reliability over complexity.

Oko utilizes local artificial intelligence to detect pedestrian signal states and deliver immediate haptic, audio, and visual feedback for blind or low-vision users. The Belgium-based team built the application from scratch during the pandemic, pivoting from hardware prototypes to Swift software while maintaining a strict focus on simplicity and real-world testing.

What is Oko and how does it function?

The application known as Oko serves as a dedicated navigation companion for individuals navigating public spaces with limited sight. The name translates directly to eye, reflecting its primary purpose of interpreting visual cues that standard users take for granted. When activated, the software continuously monitors pedestrian crossing signals through the device camera feed. It identifies whether the traffic light displays a walk or do not walk instruction and relays that information instantly. This immediate feedback loop allows users to cross streets with greater confidence and reduced hesitation.

The interface deliberately avoids cluttered menus or unnecessary settings. Instead, it provides a streamlined experience that relies on three distinct output channels. Haptic vibrations alert users to state changes without requiring them to look at the screen. Audio cues mimic familiar urban sounds to provide contextual awareness. Visual elements remain minimal but clearly indicate the current signal status when needed. This tripartite approach ensures that information reaches the user regardless of their preferred sensory input method.

Why does local processing matter for assistive technology?

The decision to run artificial intelligence models directly on the device rather than relying on cloud servers fundamentally changes how the application operates in unpredictable environments. Urban streets present constant variables such as heavy rain, blinding sunlight, wind interference, and rapidly changing traffic patterns. A network dependent system would struggle with latency or connectivity drops during critical moments. By keeping all computations within the hardware boundaries of the phone, Oko guarantees immediate response times regardless of cellular coverage.

This architecture also addresses privacy concerns that naturally accompany continuous camera usage in public spaces. Users can trust that their visual data never leaves their device to be stored or transmitted elsewhere. The technical foundation relies on a Core Machine Learning framework that processes image feeds locally. Developers originally trained the initial model using Python before converting it for seamless integration into the mobile environment. This conversion process required careful optimization to maintain accuracy without draining battery life or overheating the processor.

The Pivot from Hardware to Software

Before Oko became a polished mobile application, it existed as a cumbersome experimental prototype built during early twenty twenty one. The founding team constructed a makeshift device using microcomputers, three dimensional printed components, and borrowed audio hardware. They attached a camera to this portable rig and carried it through city streets to observe how machine learning could track traffic lights in real time.

These initial walks revealed both the potential of the technology and its physical limitations. The bulky equipment proved difficult to carry during extended outings and failed to provide the intuitive experience necessary for daily use. Recognizing that software offered a far more practical solution, the developers abandoned the hardware approach entirely. They quit their regular employment positions to dedicate themselves fully to mobile development.

How did a team without iOS experience build a reliable tool?

Constructing a functional assistive application from scratch demands rapid skill acquisition and disciplined problem solving. The three founders approached Swift development with the same analytical mindset they applied to artificial intelligence research. They consumed tutorials, watched technical videos, and relied heavily on documentation to bridge their knowledge gaps. Early coding sessions involved navigating unfamiliar frameworks while simultaneously designing user interfaces that met accessibility standards.

One founder noted opening the integrated development environment only a handful of times before beginning the project. This steep learning curve required patience and systematic debugging strategies. The team quickly discovered that standard mobile design practices often clash with assistive technology requirements. Interfaces must remain unobstructed by default system overlays or hidden menus. Controls need to be discoverable through screen readers and tactile exploration.

Navigating the Learning Curve and Early Testing

Real world validation proved essential for refining the application before public release. The founders partnered with accessibility organizations across Belgium to assemble a testing group comprising roughly one hundred participants. These volunteers provided continuous feedback during daily commutes and routine outings. Their observations quickly highlighted discrepancies between developer assumptions and actual user needs.

Most testers strongly preferred holding their devices in portrait orientation rather than landscape mode. This preference required a complete redesign of the interface layout and sensor alignment. Audio cues also needed adjustment to match familiar urban acoustics rather than synthetic tones. Visual feedback elements were expanded to ensure clarity during high traffic periods.

What does simplicity mean in accessibility design?

The philosophy of minimalism guides every aspect of the application architecture and user experience. Developers intentionally avoided adding complex features that could overwhelm users or introduce unnecessary dependencies. Early expansion ideas included tools for locating public transit stops or assisting with boarding procedures. These concepts were ultimately discarded because they threatened to complicate the core navigation function.

Assistive technology often fails when it attempts to solve too many problems simultaneously. Users require predictable, consistent interactions that reduce cognitive load during stressful situations like crossing busy intersections. The team recognized that maintaining focus on pedestrian signals would yield the highest reliability and user satisfaction. This restraint required discipline as external requests for additional capabilities continued to arrive.

The founders prioritized stability over feature accumulation by hiring specialized developers to handle advanced code optimization while they managed business operations and expansion strategy. The resulting application remains a focused tool that executes its primary function with precision. It demonstrates how deliberate constraint can produce superior outcomes in assistive software development. For broader context on industry recognition, the team's work aligns closely with the principles highlighted in introducing-the-2024-apple-design-award-finalists-24052.

The trajectory of Oko illustrates how targeted innovation can address specific mobility barriers without requiring massive infrastructure changes. By combining on device machine learning with rigorous real world testing, the developers created a system that adapts to unpredictable urban environments.

The project highlights the importance of community collaboration in assistive technology development. Direct input from individuals with visual impairments shaped every technical decision and interface adjustment. The application continues to evolve as the team expands its capabilities while preserving its foundational commitment to clarity. Future iterations will likely refine signal recognition accuracy and improve battery efficiency through ongoing optimization.

The broader implications extend beyond individual navigation to urban planning and public space design. Tools that successfully interpret traffic signals can inform how cities structure pedestrian infrastructure for maximum accessibility. As artificial intelligence matures, applications like this demonstrate that practical solutions emerge from focused development rather than broad feature accumulation.