Ideology & Principles
Key engineering decisions and the foundational principles behind sensory substitution for assistive wearables.
When a visually impaired user walks toward an obstacle at standard pace (~1.2 m/s), safety is dictated by milliseconds, not model parameters. EchoSight strictly separates reflex actions from semantic reasoning:
- Reflex Tier (<40ms): On-chip YOLOv8n + Depth ROI directly calculates spatial vector and pulses the forehead haptics without touching host CPU or cloud APIs.
- Reasoning Tier (300-500ms): SayCan task planning handles semantic interpretation ("Find an open seat") episodically via LLMs.
Audio is the primary sensory channel visually impaired individuals use to navigate the world (hearing traffic, ambient echoes, approaching people). Continuous audio chatter or spatial audio beeps create sensory overload and mask critical real-world cues.
EchoSight substitutes spatial vision with localized tactile taps on the forehead, leaving ears completely unobstructed and using open-ear bone conduction only for intentional confirmations.
Standard RGB computer vision algorithms degrade drastically on textureless surfaces (smooth indoor white walls, freshly polished tile floors) and in low-light environments.
EchoSight utilizes Active IR structured dot projection to paint an invisible mesh onto the environment, ensuring robust stereo disparity and SLAM feature tracking across all lighting conditions.
Front-heavy glasses slide down the nose bridge, cause facial fatigue, and shake during normal stride. EchoSight mounts optics and sensors on the front brow while routing compute (Pi 5) and battery to the rear cradle of a rigid halo strap to achieve a perfect center of gravity.