Table of Contents

Keynote: Hybrid Systems and Control?

Prof. Georgios Fainekos, Toyota

  • Achieve Shibuya scramble, robots moving through crowds
  • Robot needs to learn behavior of crowds, aka ML
  • Can the risk be bounded to a desirable level?
  • Modern ICCPS is heavily focused on GPU use at the edge
  • Able to apply control theory for obstacle avoidance
    • CVaR-BF
    • Performed better than reinforcement learning alone
  • Even humans struggle with pushing a cart (tractor trailer problem)
  • Hybrid A star is “classical” but modern method for global path planning
  • Proposing, safe model predictive diffusion
    • Expands to make it safe, the safety is baked into the diffusion
  • Basic MPPI, seminal work on control
  • Can get better results by applying controls to ML methods
  • Have implemented auto parking methods with a trailer
  • Toyota looks for inters every year in December
    • No formal process, email in December
  • Adversarial training through environmnet optimization
    • With Guassian splatting you are able to project objects into environment
    • Using it to find the “most” difficult configurations
    • Can train adversarial agents
  • Able to have zero shot sim to real deployment
  • Future goals of Toyota
    • FRD (Future mobility)
    • Designing robots for deployment in hospitals

SenSys: Morning sesoin

Skipped to practice presentation

SenSys: Protection & Performance: Keeping it Secure

EMPalm: Exfiltrating Palm Biometric Data via Electromagnetic Side-Channel

Haowen Xu (Worcester Polytechnic Institute); Tianya Zhao, Xuyu Wang (Florida International University); Lei Ma, Jun Dai, Alexander Wyglinski, Xiaoyan Sun (Worcester Polytechnic Institute)

Missed.

Jailbreaking Embodied LLMs via Action-Level Manipulation

Xinyu Huang (The Hong Kong Polytechnic University); Qiang Yang (University of Cambridge); Leming Shen, Zijing Ma, Yuanqing Zheng (The Hong Kong Polytechnic University)

Missed.

On Securing the Software Development Lifecycle in IoT RISC-V Trusted Execution Environments

Annika Wilde (Ruhr University Bochum); Samira Briongos (NEC Laboratories Europe); Claudio Soriente (GMV Spain); Ghassan Karame (Ruhr University Bochum)

  • Keyfort
  • Secure update and migration support
  • State continuity
  • Secure enclave local time
  • RISC-V TEE
  • Have source code online

Resolving Energy Storage for Intermittent Inference

Rei Barjami, Antonio Miele (Politecnico di Milano); Luca Mottola (Politecnico di Milano, Uppsala University, RI.SE)

  • LEACS
  • Dynamically selecting the capacitor
  • Bigger capacitor has higher leakage
  • Uses pre-recorded traces
  • Searches through possible capacitors
  • Used to increase the throughput of mode inference
  • Hit negligible benefits at 5 capacitors

R2DShield: Robust Object Detection in Real-Time via Bayesian Input Shielding

Shuo Huai, Zhixin Xie, Jun Luo (Nanyang Technological University)

  • Using NN to pre-process image to generate a purified image

Characterizing Security and Privacy Risks in Smart Home IoT Device Access Sharing

Yinxin Wan, Ting Xu (University of Massachusetts Boston); Tran Ngoc Bao Huynh, Jun Dai, Xiaoyan Sun (Worcester Polytechnic Institute); Kuai Xu, Guoliang Xue (Arizona State University)

  • Addressing access sharing (ie. txt, email, qr code) insecurity
  • Finding
    • Coarse grained access policy
    • Weak sharing controls
    • Missing/weak credentials
    • Inability to revoke device access
    • Inability to revoke sharing invitations
    • Lack of notifications for acceptance
    • Uncontrolled re-sharing
    • Over-privileged access
    • Unintended privacy exposure

On the Energy Cost of Post-Quantum Key Establishment in Wireless Low-Power Personal Area Networks

Tao Liu, Gowri Ramachandran, Raja Jurdak (Queensland University of Technology)

  • Targeting non TCP/UDP applications
  • Keys must be transported over a smalled share linka
  • Created a post-quantum key exchange mechanism for BLE
  • Link layer communication has high energy overhead
  • Post quantum is double or more the energy consumption compared to no compression

RAN-Aware Delay Compensation for Delay-Sensitive Protocols in Cellular Networks

Yuxin Liu, Tianyang Zhang, Qiang Wu (University at Buffalo SUNY); Ju Ren (Tsinghua University); Kyle Jamieson (Princeton University); Yaxiong Xie (University at Buffalo SUNY)

  • There are internal buffers inside of RAN
    • Retransmission
    • Uplink scheduling
  • Userspace tools for collecting diagnostic information in android userspace
  • Created CellNinja for mobile network diagnostic tool
  • Able to support real-time message delivery
  • GitHub repositories available

Decoupling Timing Estimation and Convergence for Efficient and Accurate Exploration of Configurations in Edge DDL Training

Joshua Daley, Pengzhan Hao, Yifan Zhang (Binghamton University)

  • Distributed method for training NN
  • Shared weights between individual workers
  • Single machine evaluation for distributed system

(Interesting) Short Paper: Simulation-Based Performance Characterization of Common NOR Flash File Systems

Yannick Loeck (Hamburg University of Technology); Christian Dietrich (TU Braunschweig); Ulf Kulau (Hamburg University of Technology)

  • Based on SpacePatch
    • Runs on RIOT OS
  • Provide simulator for testing various FS and flash types
  • VFlashSim
  • LittleFS has limitation on write speed
  • SPIFFS is very slow
  • https://github.com/2ck/flash-playground
  • Coffee file system is a possible alternative

ApproxBit: Efficient Video Analytics through Latency-Aware Offloading with Learned Binary Codes

Hyunseung Kim, Sheetal Prasanna (Purdue University); Yin Li (University of Wisconsin-Madison); Somali Chaterji ((Purdue University); Saurabh Bagchi (Purdue University, KeyByte)

  • NN based compression
  • Optimize the interface to move computation to the server
  • Fixed resolution input

MoViD: View-Invariant 3D Human Pose Estimation via Motion-View Disentanglement

Yejia Liu, Hengle Jiang, Haoxian Liu, Runxi Huang, Xiaomin Ouyang (Hong Kong University of Science and Technology)

  • Trying to address when the viewpoint changes, ie on moving orbot
  • Has human position datasets

(Interesting) TIRA: Task-Based Intermittent Remote Attestation

Fatemeh Arkannezhad, Nader Sehatbakhsh (UCLA)

  • Video presentation
  • https://ssysarch.ee.ucla.edu
  • Aimed at verifying in the integrity of onboard code for embedded systems
  • Partial task attestation at the start of ecah task
  • 18/4 memory/code overhead

SenSys: Garden Parties & Happy Smart Cities

(Interesting) SoilX: Calibration-Free Comprehensive Soil Sensing through Contrastive Cross-Component Learning

Kang Yang, Yuanlin Yang, Yuning Chen, Sikai Yang (University of California, Merced); Xinyu Zhang (University of California, San Diego), Wan Du (University of California, Merced)

  • 60B worth of produce in a year
  • Wireless sensor in the ground with antennas in the ground
    • Very strict vertial constraints
  • SoilCares, measures nutrients in the soil
  • SCARF, other systems
  • LoRa based wireless sensing, calibration free
  • Use LoRa + VNIR data to NN for soil parameters
  • There are other commercial soil moisture sensors

FlourSpec: Cost-Effective Spectral Analysis for Trace-Level Detection of Flour Adulteration

Shanwen Chen, Haiyan Hu, Yinan Zhu, Qian Zhang (HKUST)

  • Difficult to detect with just spectroscopy
  • Using LED-array spectrometer
  • NN methods for spectral reconstruction and classification

FruitScope: A Non-Invasive Fruit Ripeness Sensing System via Multi-Resolution FMCW Design and Acoustic Sensing

Shanmu Wang, Omid Abari (University of California, Los Angeles)

  • Combine mmWave and acoustic sensing
  • Take different range bins to increase resolution
  • Ground truth dataset is collected sparsely
  • Prior work, Wi-Fruit

Short Paper: Counting Parked Bicycles on the Edge - A TinyML Smart City Application

Jan Stenkamp, Mathis Hunke, Cem Karatas, Steffen Kirchhoff, Christoph Knaden, Paul Naebers, Lige Zhao, Benjamin Karic (University of Münster); Fabian Gieseke (University of Münster, University of Copenhagen); Nina Herrmann (University of Münster)

  • 8 MB Flash and RAM
  • No clue on battery lifetime

ENTS: Experiences in Co-Designed Environmental Sensing

John Madden (University of California Santa Cruz); Stephen Taylor (University of California San Diego); Laura Jaliff (Northwestern University); Aaron Wu (University of California San Diego); Alec Levy, Jack Lin (University of California Santa Cruz); Ahmed Falah (Alexandria University); Tyler Potyondy (University of California San Diego); Josiah Hester (Georgia Institute of Technology); George Wells (Northwestern University); Yaman Sangar (Georgia Institute of Technology); Pat Pannuto (University of California San Diego); Colleen Josephson (University of California Santa Cruz)

Me

MoiréLens: Bringing Schlieren Imaging into Real-World Environments Using Moiré Patterns

Linzhen Zhu, Runqiu Wang, Yi Rong, Ke Sun (University of Michigan Ann Arbor)

  • Schlieren image detects changes in gas concentration
  • Moire is the interaction between a display and CMOS camera
  • Want to detect airflow visualization
  • Issues with alignment between the screen and camera
  • Increased resolution gives greater distance
  • https://ambient-intelligence-lab-umich-eecs…

MagTach: Non-Intrusive Long-Range Magnetic Tachometry via Harmonic-Aware Inductive Sensing

Rahul Sidramappa Hoskeri, Hua Huang (University of California Merced)

  • Magnetometer is main contribution, have limited range
  • Inductive sensor array for higher order harmonics
  • Filtering amplified the harmonics
  • NN methods for extracting harmonics
  • Accurate up to 1m
  • Up to 10kHz

VehicleSense: Adaptive Sensor-Video Learning for Vehicle Type Identification and Overload Detection

Ashutosh Kr Sinha, Sushant Dagaji Desale, Nirjay Kumar, Rahul Mishra (IIT Patna)

  • Acceleration sensor to record vibration signature
  • NN method to determine class probabilities

SPIDER: Lightweight Speaker Identification on Resource-Constrained Embedded Devices

Markus Gallacher, Carlo Alberto Boano (Graz University of Technology); Arun Sankar Muttathu Sivasankara Pillai (South East Technological University); Utz Roedig (University College Cork); Willian Lunardi, Michael Baddeley (Technology Innovation Institute)

  • https://open-earable.teco.edu
  • Models are too big for small mcu
  • Don’t want to offload to the cloud
  • Can “compound scale” the NN
  • Fundamentally a way to scale down NN models for speech recondition
  • For an NN ensemble, only take high confidence segments

RippleSense: Scalable and Efficient Wideband Spectrum Sensing

Andreas Kuster, Yanbo Zhang (Nanyang Technological University); Mo Li (Hong Kong University of Science and Technology)

Had to skip to catch the bus

From Cheap to Chic: Enhancing Music Playback Quality of Budget Earphones via Hardware-Aware Learning

Changshuo Hu, Hung Manh Pham (Singapore Management University); Ting Dang (The University of Melbourne); Jiannan Li, Rajesh Balan (Singapore Management University); Dong Ma (University of Cambridge)

Had to skip to catch the bus