Track: AI Meets the Physical World

9:00am - 9:40am

Deep Learning on Microcontrollers

Learn why deep learning is a great fit for tiny, cheap devices, what you can build with it, and how to get started.

10:00am - 10:40am

From Robot Simulation to the Real-World

Simulation is one of the most powerful tools in the robot developer's tool belt. Besides allowing quicker, safer and cheaper iterations, it can be used to prototype before building, run continuous integration, train machine learning algorithms, etc. One popular open source robotics simulator is Gazebo, maintained by Open Robotics, the same foundation that maintains ROS, the Robot Operating System. Combined, ROS and Gazebo are used by an increasing number of developers around the world.

Gazebo's development started over 17 years ago, but its most current form started taking shape in 2012, when DARPA sponsored Open Robotics to run the Virtual Robotics Challenge, the first stage of its Robotics Challenge. A total of 26 teams competed in the virtual competition, controlling an Atlas robot from Boston Dynamics in a simulated disaster scenario. As a result of the virtual competition, the top 7 teams received funding to compete in the final competition with the same robot, but this time a physical one in a physical scenario.

Ever since, Gazebo has continued to be developed and improved to better support various types of robots, spanning ground, water and air, and is being increasingly used by the academia, industries and, sure enough, in other competitions. In this talk, Louise will give an overview of Gazebo's architecture and go over some examples of projects using Gazebo which Open Robotics has been involved with, describing how they bridged virtual robots to their physical counterparts.

11:00am - 11:40am

The Road to Artificial Intelligence: An Ethical Minefield

There is no doubt that developments in artificial intelligence offer significant benefits to humanity. However, are we properly considering some of the negative externalities that could accrue to society? This presentation offers a robust look into the complex ethical issues faced by today's top engineers and poses open-ended questions for the consideration of attendees. It places special focus on the rise of autonomous vehicles and their potential susceptibility to attacks by malicious agents, while also covering adversarial intrusions into machine learning engines more broadly.

12:00pm - 12:40pm

DeepRacer and DeepLens, Machine Learning for Fun! (and Profit?)

 In 2017, Amazon announced the DeepLens, a machine learning enabled camera, which they released in 2018.  In 2018 they announced the DeepRacer, a 1/8 size model race car that is basically a DeepLens on wheels, which they will release in July.  In this talk, you'll hear about the speaker's attempts to do cool things with these machine learning "toys", learn about some of the basics of machine learning necessary to understand what's happening in the devices, and, assuming the device isn't being finicky, see a DeepRacer in action!  We will also have a torn down DeepRacer and DeepLens for you to look at and (gently) play with.  We'll also go over some possible real life use cases for the devices.  Come join us for machine learning fun!

1:40pm - 2:20pm

Evoking Magic Realism with Augmented Reality Technology

2:40pm - 3:20pm

Advanced Topics in Autonomous Driving using Deep Learning

Autonomous vehicles need to perceive their surroundings and analyze it to make decisions and act in an environment. More specifically, the autonomous vehicle detects objects on the road and maneuver through the traffic utilizing smart functional modules. In recent years, artificial intelligence, in particular, deep neural networks, have been used widely to build these smart functional modules.

While object detection (putting bounding boxes around objects), and semantic segmentation (labeling each pixel in an image) has been the focus of many researchers in autonomous driving, these methods may fall short when it comes to forming a better social understanding of pedestrian intent. In this talk, we present our approach to pedestrian intent prediction and communication, which leverages more complex computer vision algorithms that estimate human pose rather than bounding boxes or pixel labels.

The increasingly sophisticated models, like the pose estimation networks we describe, show tremendous promise as they prove to be robust at approximating complex and non-linear mapping functions from images to outputs. However, these models are typically large and have a huge number of parameters resulting in a steep cost in terms of training and inference time resource requirements. This makes the use of these networks challenging on resource and power constrained embedded systems. In this talk, we also show that the compression of neural networks results in faster predictions with smaller deep neural networks.