A full day virtual workshop, June 4, 2021
Talks and Presentations: https://www.youtube.com/playlist?list=PLLQ5-0tmXjiJRtVHNcZ9quwhUDwV2L9Jh
Posters and Demos: https://gather.town/i/PoM0B4Uz
Almost all real-world AI and robotics applications are power-constrained and depend on real-time understanding of the robot's environment. Robotic systems that rely on simultaneous localization and mapping (SLAM) and other perception algorithms, would greatly benefit from novel emerging computing paradigms such as computation on or near the focal plane. In this workshop, we explore the emerging unconventional perception methods such as computation directly within the light sensor, performing early-stage recognition before converting signals to digital form, and implementing all or parts of the deep neural networks on the chip. In these methods, only where features of value are identified is data actually transferred to later stages of the vision pipeline. We aim to bring researchers from academia and industry from across multiple disciplines, including i) hardware design, ii) compiler, iii) software/algorithm design, and iv) holistic design to demonstrate the power of novel computational approaches using these unconventional sensors. These novel approaches enable a variety of recognition and tracking applications with exceptionally low latency, high frame rate and low power.
Several live demos using SCAMP5-d Focal-plane Sensor-processor Arrays (FPSP) will be presented. These demos are:
BIT-VO: running high-speed visual odometry on SCAMP5-d FPSP
Neural Sensor: HDR imaging using SCAMP5-d FPSP
CAIN: demonstrating a new pipeline for compiling CNN kernels to SCAMP5-d FPSP code.
AnalogNet: a robotic navigation demo using a SCAMP5-d FPSP, capable of navigation in indoor environments avoiding all obstacles, at high frame rate and low power
Focal Plane Inference: a demonstration of CNN inference performed on the focal plane of a SCAMP5-d for digit recognition.
BIT-VO
Neural Sensor
CAIN
AnalogNet
Topics of interest include, but not limited to:
Focal-plane Sensor-processor Arrays
Computational Sensors
Smart Vision Sensors
Event Cameras
Novel Depth Ranging Sensors
Compiler Methods for Emerging Sensors
Application in Robotics and AI
AI on the Focal Plane
SLAM on Chip/Focal Plane
Visual/inertial Odometry on Chip/Focal Plane
Metrics, Benchmarking, and Datasets for the Emerging Sensors
Performance Analysis of Algorithms and System using Unconventional Sensors
Imperial College London
Stanford University
University of Zurich
University of Tokyo
University of Manchester
Imperial College London
Australian National University
Sony
Talks and paper presentations are live. Each talk is 20 minutes plus 10 minutes for Q&A, right after the talk. Paper presentations are 7 minutes plus 8 minutes for Q&A, right after the presentation.
You may add the event to your calendar via this link:
https://calendar.google.com/calendar/u/0?cid=aWNyYS53b3Jrc2hvcEBnbWFpbC5jb20
Event Camera
PDF, Video, "Event-based Asynchronous Sparse Convolutional Networks," Nico Messikommer, Daniel Gehrig, Antonio Loquercio and Davide Scaramuzza
PDF, PPT, "Classification of motion directions with spiking neurons: a continual learning case," Veïs Oudjail and Jean Martinet
PDF, Video, "DSEC: A Stereo Event Camera Dataset for Driving Scenarios," Mathias Gehrig, Willem Aarents, Daniel Gehrig and Davide Scaramuzza
Focal-plane Sensor-processor Arrays (FPSP), Pixel Processor Array (PPA)
PDF, Poster, Video1, Video2, "Direct Servo Control from In-Sensor CNN Inference with A Pixel Processor Array," Yanan Liu, Jianing Chen, Laurie Bose, Piotr Dudek and Walterio Mayol-Cuevas
PDF, Poster, "Super-pixel Compilation for Pixel Processor Arrays," Edward Stow and Paul Kelly
PDF, Poster, Video, "Bringing A Robot Simulator to the SCAMP Vision System," Yanan Liu, Jianing Chen, Laurie Bose, Piotr Dudek and Walterio Mayol-Cuevas
Hardware
PDF, Poster, "Towards Dynamic Fault Tolerance for Hardware-Implemented Artificial Neural Networks: A Deep Learning Approach," Daniel Gregorek, Nils Hülsmeier and Steffen Paul
The workshop accepts contributions of research papers describing early research on emerging topics.
The workshop is intended for quick publication of work-in-progress, early results, etc. The workshop is not intended to prevent later publication of extended papers.
Prizes will be given to the best presentation and also to the best poster/video.
Submission Format: please use standard IEEE format (minimum two pages, and up to eight pages). For more information, see the IEEE website https://journals.ieeeauthorcenter.ieee.org
CFP Link: https://easychair.org/cfp/ONSVP-2021
Submission Link: https://easychair.org/conferences/?conf=onsvp2021
Full paper submission due date: April 15, 2021 April 30, 2021
Paper acceptance notification: May 18, 2021
Camera-ready version due date: May 28, 2021
Final Submission Link: https://easychair.org/conferences/?conf=onsvp2021
Workshop day: June 4th, 2021 (See the schedule above for the start time)
Youtube Recordings (talks and paper presentation):
https://www.youtube.com/playlist?list=PLLQ5-0tmXjiJRtVHNcZ9quwhUDwV2L9Jh
Gather Town Link: (For posters and demos): https://gather.town/i/PoM0B4Uz
Paul Kelly (Imperial College London, UK)
Riku Murai (Imperial College London, UK)
Yu-Hsin Chen (Facebook, USA)
Raluca Scona (Imperial College London, UK)
Laurie Bose (University of Bristol, UK)
Thomas Debrunner (iniVation AG, Switzerland)
Edward Stow (Imperial College London, UK)
Program Committee Chairs:
Julien Martel (Stanford University, USA)
Sajad Saeedi (Ryerson University, Canada)
Sajad Saeedi (Ryerson University, Canada)
Julien Martel (Stanford University, USA)
Riku Murai (Imperial College London, UK)
Paul Kelly (Imperial College London, UK)
If you have any questions, please contact organizers at icra.workshop@gmail.com
