Mat Kelcey - Deep RL for Robotics 

*** VIDEO WITH FIXED SOUND HERE: • Mat Kelcey - Deep RL f...
From the Melbourne Machine Learning and AI Meetup in September 2017 mlai.melbourne/
Talk starts at 13:20
Mat Kelcey - Deep Reinforcement Learning for Robotics
Reinforcement learning (RL) is making a comeback in robotics, driven partly by the huge recent successes in using deep neural networks for perception and control. In this talk we'll do a brief intro to RL and show how neural networks are a perfect fit for the function approximators these algorithms need. We'll go over some standard architectures and practical approaches to training them with end to end examples; i.e. pixels to motor control. We'll also review generative & domain adversarial approaches for transferring models trained in simulation to a physical robot.
Oh, Mat has a blog too:
matpalm.com/blog/