Challenges in autonomous robotic manipulation

There have been recently announced breakthroughs in artificial intelligence and robotics. AlphaGo - the Go-playing system defeated one of the best Go players, deep Q-network could learn and play video games on levels comparable to professional human game testers. On the other hand, Boston Dynamics presented a humanoid Atlas robot which can walk in difficult terrain, while Google demonstrated a robotic system based on convolutional neural networks which can autonomously learn to grasp objects. 

How far are we then from humanoid robots known from Sci-Fi films? In this presentation I am going to argue that such universal autonomous robots are not be possible to realise in any foreseeable future, however we are about to enter the era of domain-specialised intelligent robots in industry and everyday life.

I will present the kind of problems roboticists are currently struggling with, as well as some promising approaches which have been successfully applied in real-world tasks. I will mostly focus on robotic applications in domestic environments, during handling dangerous materials, grasping and packaging items at warehouses or controlling artificial limbs in prosthetics.