Deep Learning for Automated Systems: From the Warehouse to the Road - - PowerPoint PPT Presentation

▶
deep learning for automated systems from the warehouse to
SMART_READER_LITE
LIVE PREVIEW

Deep Learning for Automated Systems: From the Warehouse to the Road - - PowerPoint PPT Presentation

Sep 03, 2022 384 likes •674 views

Deep Learning for Automated Systems: From the Warehouse to the Road Dr. Melissa C. Smith Clemson University Future Computing Technologies Laboratory Eddie Weill, Sufeng Niu, Colin Targonski, and Ben Shealy Overview Simulation Using deep

slide-1
SLIDE 1

Deep Learning for Automated Systems: From the Warehouse to the Road

  • Dr. Melissa C. Smith

Clemson University Future Computing Technologies Laboratory Eddie Weill, Sufeng Niu, Colin Targonski, and Ben Shealy

slide-2
SLIDE 2

Overview

slide-3
SLIDE 3

Simulation

Using deep learning in a simulated environment

slide-4
SLIDE 4

The Need for a Simulator

slide-5
SLIDE 5

Deep Learning in a Simulator

Webpage: http://www.carla.org/ Paper: Dosovitskiy, Alexey, et al. "CARLA: An open urban driving simulator." arXiv preprint arXiv:1711.03938 (2017).

slide-6
SLIDE 6

Perception

Developing perception for automated systems

slide-7
SLIDE 7

Autonomous Driving Perception

slide-8
SLIDE 8

Autonomous Driving Perception

slide-9
SLIDE 9

Autonomous Driving Perception

Tetreault, Jesse. Deep Multimodal Fusion Networks for Semantic Segmentation. Diss. Clemson University, 2017.

slide-10
SLIDE 10

Perception on Embedded Devices

Detector Network (i.e. YOLO, SSD) OCR Text Extraction

Detections Extracted

Acknowledgement goes to BMW ITRC for partnering on this endeavor and providing the data for experimentation

Luckow, Andre, et al. "Deep learning in the automotive industry: Applications and tools." Big Data (Big Data), 2016 IEEE International Conference on. IEEE, 2016.

Synthetic Image Training

Redmon, Joseph, et al. "You only look once: Unified, real-time object detection." Proceedings of the IEEE conference on computer vision and pattern recognition. 2016.

slide-11
SLIDE 11

Perception + Control

Integrating perception with reinforcement learning

slide-12
SLIDE 12

Autonomous Driving in CARLA

Rewards State (RGB Image) Action (Steering / Acceleration)

Combine Feature Maps Environment (Carla) Segmentation Reinforcement Learning Agent Object Detection Traffic Interpretation

“Approaching Stop Sign” “Light is Yellow” “Car stopped at intersection”

slide-13
SLIDE 13

Autonomous Driving in CARLA

slide-14
SLIDE 14

Planning

Using reinforcement learning to explore environments

slide-15
SLIDE 15

Neural Network Based Planning

VIN GVIN

16 x 16 2D Maze New York City Street Map (13K intersections)

  • S. Niu, et al. Generalized Value Iteration Network: Life Beyond Lattices. 32nd AAAI, 2018.
  • A. Tamar, et al. Value Iteration Networks. NIPS, 2016.
slide-16
SLIDE 16

Neural Network Based Planning

Navigation Knowledge querying Social network reasoning Network routing

S G

slide-17
SLIDE 17

Neural Network Based Planning (GVIN)

Start Goal

slide-18
SLIDE 18

Neural Network Based Planning (GVIN)

Testing Testing Training 10 nodes 2642 nodes 5069 nodes

  • S. Niu, et al. Generalized Value Iteration Network: Life Beyond Lattices. 32nd AAAI, 2018.
slide-19
SLIDE 19

Distributed Computing

Scaling beyond a single HPC cluster

slide-20
SLIDE 20

Clemson University’s Palmetto Cluster

slide-21
SLIDE 21

Is One HPC Cluster Enough?

www.nlm.nih.gov Image credit: http://harborresearch.com/connected-vehicles-rise-transportation-ecosystems/

slide-22
SLIDE 22

Scientific Data Analysis at Scale (SciDAS)

NSF Award No. 1659300

slide-23
SLIDE 23

Scientific Data Analysis at Scale (SciDAS)

NSF Award No. 1659300

slide-24
SLIDE 24

Scientific Data Analysis at Scale (SciDAS)

NSF Award No. 1659300

slide-25
SLIDE 25

Scientific Data Analysis at Scale (SciDAS)

NSF Award No. 1659300

slide-26
SLIDE 26

Scientific Data Analysis at Scale (SciDAS)

NSF Award No. 1659300

slide-27
SLIDE 27

Wrap Up

slide-28
SLIDE 28

Thank you!

Questions?