Skip to main content
GMLPi
中文
Featured projects

Collaborative R&D · Wearables

AI Assistive Glasses

Vision-assistance glasses for people who are blind or have low vision. The project turns visible information into speech and directional cues for finding objects, reading, and confirming destinations.

Discuss this project
Watch the demo
Illustration of the glasses Watch on Bilibili

Visual perception and interaction

Scene description
When asked about the surroundings, the glasses describe visible objects, signs, and spatial cues aloud.
Voice-guided object finding
Name an object to search the visible scene. Direction and distance cues help the user approach it.
Text reading
Read recognizable text on labels, room signs, menus, and printed pages aloud.
Travel cues
Combine route direction and the visible surroundings with short spoken and directional haptic cues to help users confirm heading and destination.

Hardware development

From feasibility analysis to prototype retest

Evaluation models, design views, and bench records from different prototype revisions.

  1. Feasibility analysis

    Start with reading, object finding, and travel cues. Break down response time and check power headroom within the compact frame.

    Two feasibility models. End-to-end latency is the sum of capture, processing, and feedback times. Loop voltage drop equals current times loop resistance. Resistive loss equals current squared times loop resistance.
    Latency breakdown and power-loop model
    End-to-end response
    Account for capture, processing, and feedback from image acquisition to a spoken or haptic cue. Measure the full path while concurrent tasks are running.
    Voltage drop and line loss
    Estimate voltage drop and resistive loss through the flexible interconnect using load current and loop resistance. Check power headroom alongside fit constraints and thermal observations.

  2. Circuit design and PCB routing

    Translate the circuit design into component placement and routing on a narrow board.

    Schematic excerpt of the prototype controller and peripheral circuits.
    Schematic excerpt
    Pads, vias, and multilayer routing in the PCB editor.
    PCB routing view

  3. Board shape and assembly review

    Check board edges and pad positions in 2D, then use the 3D model to review component and connector clearance.

    Top view of the blue PCB showing its outline and gold pads.
    2D PCB view
    3D PCB model with components and connectors in place.
    3D PCB assembly model

  4. Assembly inspection and sensor bring-up

    Inspect the assembled board under a microscope, then connect it to the test bench to check sensor data.

    Microscope view of connector pins and nearby surface-mounted components.
    Connector and components under magnification
    Microscope view of surface-mounted components and solder joints.
    Surface-mount assembly close-up
    Prototype board and flex cable in front of a display showing sensor traces.
    Prototype sensor bring-up

  5. Waveform and thermal checks

    Check relative channel timing on the bench, then use thermal imaging to observe temperature distribution across the prototype.

    A prototype board connected to oscilloscope probes, with several rising channel traces on the screen.
    Multi-channel oscilloscope capture View multiple test points together to compare rising edges, relative timing, and settling.
    Thermal capture 01, readings of 55.7°C and 29.7°C Thermal capture 02, readings of 34.0°C and 29.8°C Thermal capture 03, readings of 47.4°C and 28.2°C Thermal capture 04, 39.1°C near the flex cable and 29.6°C at the other point
    Four prototype thermograms Observe temperature patterns around the board and flex cable. Open each image to read the marked values for that test.

  6. Fault isolation and retest

    In one prototype bring-up, the team traced abnormal image data to solder joints and reran the complete imaging workflow after rework.

    1. Compare captured dataUse reference data to identify a repeatable imaging fault.
    2. Inspect solder jointsInspect the connector area associated with the affected signals under a microscope.
    3. Retest after reworkRerun the imaging workflow and confirm normal operation on the retested boards.

Typical use cases

A person with a white cane stops near an entrance sign at a public building.
Illustrative scenario · Confirming the entrance to a public building

Finding everyday objects indoors

For requests such as “find my keys” or “find the remote,” report whether the object is visible and guide the search direction.

Reading signs and printed text

Read text from the current view when identifying a room number, checking a label, or reading a book.

Confirming the final approach

Once navigation reaches the vicinity, use room numbers, storefront signs, or entrance markers to help confirm the final location.

Evaluation in real environments

Image quality and task outcomes
Check whether text is readable, targets are recognizable, and directional cues are clear in low indoor light, outdoor backlighting, and motion while walking.
Speech and haptics
While walking, evaluate the clarity of short voice and directional haptic cues. For reading, focus on intelligibility and ease of interrupting playback.
Handling uncertainty
When the image is blurred, the target is blocked, or connectivity is unstable, explain what can be confirmed and when to capture another view. Users can combine travel cues with their existing mobility aids and judgment.