Last Week in AI #111

U.S. AI workforce, using RL to train bipedal robots, and more!

Mini Briefs

Analysis on U.S. AI Workforce

In a new report published by Georgetown University's Center for Security and Emerging Technology, researcher analyze recent trends in the U.S. AI workforce and found that the field is experiencing rapid growth. The number of U.S. AI workers grew more than 20% from 2015 to 2019, compared to just 5.6% for all U.S. occupations. This growth seems to be especially concentrated in the top 5 urban hubs - Boston, NYC, San Francisco, Seattle, and Washington D.C. While there is strong evidence that growth will conintue, it will probably occur at a slower pace, estimated to be at 8% over a 10yr period from 2019-2029.

Forget Boston Dynamics. This robot taught itself to walk

A new paper from UC Berkeley demonstrated using Reinforcement Learning (RL) to teach a biped robot how to walk and robustly handle external perturbations. The robot is trained entirely in simulation, and its specific algorithm design allowed the learned agent to exhibit a wide range of behaviors without hardcoding. In addition to normal walking, other behaviors include changing walking heights and pushing out one leg to recover from external pushes. This is different from the impressive Boston Dynamics demos with the biped robot Atlas, because in that case the robot controllers were all hand designed.

While the work is impressive, MIT Tech Review's characterization as this is the first time ever that RL has been used to train a biped robot is misleading. There is a long history of applying RL in robot locomotion, and RL for bipeds has been introduced since at least 1997.

Podcast

Check out our weekly podcast covering these stories! 
Website | RSS | iTunes | Spotify | YouTube

News

Advances & Business

Concerns & Hype

Explainers


That's all for this week! If you are not subscribed and liked this, feel free to subscribe below!

Recent Articles:

Copyright © 2021 Skynet Today, All rights reserved.