View
1
Download
0
Category
Preview:
Citation preview
Artificial intelligence for self-driving cars2016. 06. 03.
arpad.takacs@adasworks.com
2016. 06. 03. 1Árpád Takács: AI for self driving cars
AdasWorks revolutionizes the automotive industry by
building artificial intelligence software for automated driving.
2016. 06. 03. 2Árpád Takács: AI for self driving cars
Founded in July, 2015
AdasWorks is a spin-off of Kishonti Ltd.
High-performance embedded programming
Computer Vision
Mapping and navigation
Robotics
Automotive engineering
Khronos Workgroup
Embedded Vision Alliance
Professional partners
2016. 06. 03. Árpád Takács: AI for self driving cars 3
Artificial Intelligence
Neural Networks
Unsupervised learning
Reinforcement learning
Multipe camera support
Integrated hardware agnostic
Scalable and future proof
Quick testing and validation cycle (GPU-based)
2016. 06. 03. 4Árpád Takács: AI for self driving cars
Dr. Richard Satava,
University of Washington
2016. 06. 03. Árpád Takács: AI for self driving cars 5
ADAS – Advanced Driver Assistance Systems
Goal: increasing driving safety and enhancing driving experience
Automation – Adaptation – Enhancement
Cameron Gulbransen Kids Transportation Safety Act
2016. 06. 03. 6Árpád Takács: AI for self driving cars
2016. 06. 03. Árpád Takács: AI for self driving cars 7
2016. 06. 03. 8Árpád Takács: AI for self driving cars
2016. 06. 03. Árpád Takács: AI for self driving cars 9
2016. 06. 03. Árpád Takács: AI for self driving cars 10
2016. 06. 03. Árpád Takács: AI for self driving cars 11
2016. 06. 03. Árpád Takács: AI for self driving cars 12
2016. 06. 03. Árpád Takács: AI for self driving cars 13
2016. 06. 03. Árpád Takács: AI for self driving cars 14
Deep Neural Network library
Standard routines implemented and accelerated
Highly supported by DNN frameworks
BLAS library for CUDA
Fully connected networks
Accelerated FFT library for CUDA
General image processing
2016. 06. 03. Árpád Takács: AI for self driving cars 15
Covering 70-80% of the use cases
Supported and continously developed kernels
Extendable
Special datasets
Extreme use cases
Workload
2016. 06. 03. Árpád Takács: AI for self driving cars 16
Per pixel operation
Image filtering
Optical flow: per pixel, per feature
Detection: HOG, LBP, VJ
SLAM
Hybrid solutions
Parameter setting, parameter learning
3D bounding box detection with SFM
Production and development
2016. 06. 03. Árpád Takács: AI for self driving cars 17
2016. 06. 03. Árpád Takács: AI for self driving cars 18
2016. 06. 03. Árpád Takács: AI for self driving cars 19
2016. 06. 03. 20Árpád Takács: AI for self driving cars
2016. 06. 03. Árpád Takács: AI for self driving cars 21
2016. 06. 03. Árpád Takács: AI for self driving cars 23
2016. 06. 03. 24Árpád Takács: AI for self driving cars
2016. 06. 03. 26Árpád Takács: AI for self driving cars
2016. 06. 03. 27Árpád Takács: AI for self driving cars
Incorrect reconstruction of the environment
Incorrect decision making and reaction
Unknown/unlearned situations
Value of human life
Legal frameworks
Home-made software
2016. 06. 03. 28Árpád Takács: AI for self driving cars
Håkan Samuelsson
CEO - Volvo Cars
2016. 06. 03. Árpád Takács: AI for self driving cars 29
2016. 06. 03. Árpád Takács: AI for self driving cars 30
Recommended