Second day of the conferences. The first section had some interesting papers on UWB radar sensing. Both UWB and mmWave are booming with the advances in NN able to process the complex data streams. I skipped the second session as it mainly pretains to AI, instead opting to go to the RTAS session on controls which was far more mathmatically rigious than I expected. There was some interesting applications relating to control and scheduling that could be useful to scheduling packets for WSNs. Unfortunately, I had to head home so I missed the last session and will be missing tomorrow.

SenSys

UMusic: UWB radar for in-car occupancy sensing (Bosch)

  • UWB Ultra wide band
  • In-car occupancy for safety and security
  • Can be a replacement for keyless entry
    • 45% of all new vehicles to support UWB
  • CarOSense: ML for occupancy prediction
    • RF channel is noisy and env dependent
    • Need to get a lot of data, hard to verify
  • Accuracy of 90%, improves previous work by 15%
  • Cost is the main drive for the number of sensors

RAM-Hand: Robust Acoustic Multi-Hand Pose reconstruction using microphone

array (Shiyang Wang)

  • Applications: Smart home, Gesture control, VR
  • main contribution multi-hand tracking
  • Beyond-Voice is previous work

LeakerFeeder: In-Air Gesture control through leaky acoustic waves (Yongjie Yang)

  • Example of leaky sounds is headphones
  • Measures how far you hand is away from the headphones
  • Uses feed forward microphones found in modern airbuds
  • FMCW SONAR??
  • Leverages how gestures are performed
  • Uses NN to predict accurate finger positions
  • 89% accuracy to detect gestures

LightLLM: A versatile LLM for predictive light sensing (Jiawei Hu)

  • PLS Predictive Light Sensing
  • Light sensors can be used for indoor localization
  • Pretrained model for light based sensing that can be adapted to tasks
  • Very LLM heavy
  • cGAN is alternative to LightLLM, gets outperformed by x4.4

RoboTera: Non-contact friction sensing for robotic grasping … (Vahid

Yazdnian)

  • Knowing friction can help choose where to grasp an object
  • Existing solutions are tactile-based sensing
  • Uses a Sub-Tz radar to precept friction
  • Capturing the roughness of the object to determine friction
    • Roughness if the RMS of the distance signal
  • Identify material using an onboard RGB camera with CNN
  • Calculating the area where the gripper will grab
  • Can handle ceramic but no comment on glass

LiDARMarker: Machine-friendly Road markers for Smart Driving Systems

(Fengxu Yang)

  • ADS Autonomous Driving Systems
  • Current road markers are not machine friendly
  • Address both human and ADS readability
  • Uses LiDAR to mark road signs
    • Assumes that ADS uses LiDAR
  • Apply infrared-absorbing paint
  • QR codes are difficult to implement
  • Uses barcode-inspired, fan shaped layout
    • Similar to a WiFi icon
  • Reliable decoding up to 15 m
  • Does the size of the marker affect the range?

RTAS

I have a lot less experience with the current SOTA in control systems so the notes for this section are more sparse.

CP-Sensi-based scheduling for mesh networks

  • Missed first part…
  • Scheduling for cyber physical systems for control problems
  • Scale up to large amount of devices (0.6s for 500 devices)
  • Uses 802.15.4 mesh networking
  • Adjust priorites based on multi-time-granularity control performance and network conditions
  • Used TOSSIM
    • Possibly relevant to our work with LoRaWAN

Analysis of Control Systems under Sensor Timing Misalignment

  • Example, skip in a camera frame can cause sensor fusion errors
  • Constructs the theory behind if a sensor becomes misaligned
  • Process yields stability analysis of delay of each sensor

Scheduling EV battery swap (Jaeheon Kwak)

  • Swapping batteries instead of charging packs
  • Assume that a replacement is 3-10 minutes
  • Mainly addressing the charge time, need to be constant
  • Control to guarantee battery is swapped in time
  • quasi-non-preemtive dual-priority-fifo