Trajectory and Sway Prediction for Fall Prevention

Position

Stanford ARMLab Researcher

Collaborators

Ken Wang (PhD), Monroe Kennedy III (PI)

Responsibilities

Develop an iOS application for data collection and depth panorama generation

Skills

  • Swift
  • Python

Challenge

  • Conducted research with the Stanford Assistive Robotics and Manipulation Laboratory (ARMLab) to develop a wearable sensor that can predict a person’s path and stability, then alert them of possibile falls in real time.
    • Intended users are older individuals who are at higher risk of instability and falls.
  • Prior data collection setup involved multiple pieces of hardware. I aimed to simplify the setup by reducing the number of parts while maintaining performance.

Action

  • Developed an iOS application in Swift (programming language) that leverages the LiDAR sensor in iPhones to collect environment data and generate a depth panorama
    • The depth panorama is used as input to a VAE + LSTM trajectory and sway prediction model

Result

  • Application demo (below) shows the generation of the depth panorama in real time as a person walks down a hallway. The FPS on the screen needs to be multiplied by the number of GPUs (6 for the iPhone 12 Pro that the testing was conducted on)
  • PyTorch prediction model has been converted to CoreML and will eventually be deployed on the iPhone for an integrated data collection and prediction process

Research Presentations

  • SURI Poster Session, August 2023: Presented summer research progress to Stanford mechanical engineering labs and the SURI cohort
  • Innovation and Discovery Expo, October 2023: Presented to biosciences, engineering, and medical researchers as part of the ARMLab demonstrational booth. The expo was hosted by Stanford Bio-X and Wu Tsai Human Performance Alliance.

Research Poster