Pathlight Wearable Navigation Prototype

Historical engineering snapshot: this repository documents Pathlight's earlier exploration of a wearable navigation system for blind and low-vision users. The project has since changed direction; the current Pathlight product is a voice-first email and calendar assistant described at pathlightproj.com.

This codebase explored an on-device perception pipeline for a Jetson-based headset: stereo cameras estimate depth, object detection identifies potential obstacles, a path planner selects a direction, and audio or LEDs communicate system output. It also contains experiments in face memory, voice interaction, and hardware control.

The repository is useful as a record of the system decomposition and deployment research. It is not a complete, validated mobility aid and must not be relied on for personal safety.

Prototype scope

Area	Approach represented in the repository
Perception	YOLOv8 object detection plus OpenCV stereo disparity and depth estimation
Navigation	Grid-based obstacle mapping, path scoring, and 2D/3D plan combination
Social memory	Face encodings and interaction records stored locally
Interaction	Text-to-speech, speech recognition, LED direction output, and an assistant adapter
Hardware	Abstractions for dual IMX219 cameras, an IMU, LEDs, audio, and a microcontroller
Target platform	NVIDIA Jetson Orin Nano with JetPack 6.2.1 / CUDA 12.6-era setup scripts

Some hardware adapters fall back to simulated data when Jetson libraries or peripherals are unavailable. Those fallbacks are development scaffolding, not evidence of end-to-end hardware operation.

System shape

dual cameras
    │
    ├──> stereo depth ──> 3D obstacle candidates ──┐
    │                                              ├──> path selection
    └──> object detection ──> 2D obstacle map ─────┘         │
                                                             ├──> LED direction
microphone ──> speech recognition ──> assistant ─────────────└──> spoken feedback

camera ──> face recognition ──> local interaction memory

.
├── core/
│   ├── vision/       # Detection, face recognition, and stereo depth
│   ├── navigation/   # Path planning
│   ├── memory/       # Local face and interaction records
│   └── audio/        # Audio-processing experiments
├── hardware/         # Camera, LED, IMU, audio, and microcontroller adapters
├── ai/               # Assistant and voice-processing experiments
├── scripts/          # Jetson provisioning, diagnostics, and systemd units
├── docs/             # Hardware-specific setup and development notes
├── config/           # Example runtime configuration
└── main.py           # Intended orchestration entry point

Reading and reproducing the work

The setup material is hardware-specific and assumes familiarity with Jetson recovery, JetPack, CUDA-enabled OpenCV, and source-built dlib. Start with:

1. PATHLIGHT_FINAL_SPECIFICATIONS.md for the recorded target environment.
2. COMPLETE_JETSON_SETUP_GUIDE.md for the installation sequence.
3. config/config.example.yaml for the expected runtime configuration.
4. scripts/test_installation.py for environment diagnostics.
5. docs/stereo_vision_guide.md for calibration and depth-pipeline notes.

Do not place API credentials in tracked configuration. Copy the example configuration to an ignored local file and supply credentials only in your own environment.

Known limitations of this snapshot

This repository is preserved as-is rather than presented as a finished release:

• main.py does not currently pass Python compilation because of indentation errors.
• The intended AudioIO integration is referenced by the orchestrator but its import is disabled.
• The assistant adapter uses the pre-v1 OpenAI chat-completions interface while requirements.txt permits newer client versions.
• Camera calibration files, model weights, and device-specific runtime data are not included.
• Installation and stereo tests require the target Jetson hardware and were not rerun as part of this documentation pass.
• Several setup guides overlap because they capture successive deployment iterations.

Safety, privacy, and data handling

A vision-and-memory prototype can process faces, voice, location context, and environmental imagery. Any future use should include explicit consent, retention controls, encryption, threat modeling, accessible failure states, and supervised field validation. This repository does not establish those safeguards or any medical, accessibility, or safety certification.

Project status

Development in this repository stopped in 2025. It remains public to document an early assistive-computing architecture and the engineering constraints of deploying computer-vision workloads on embedded hardware.

No license file is included; absent an explicit license, the source is not offered for reuse or redistribution.