Neural SLAM: Learning to Explore with External Memory

We provide a way for agents to learn the expression of global map from sensor data and help them explore in the new environment. In order to achieve this goal, we incorporate a process that mimics the conventional simultaneous positioning / mapping (SLAM) to soft attention based addressing of the external memory architecture, where external memory serves as the internal representation of the environment. This structure encourages the evolution of SLAM-like behavior in fully differentiable deep neural networks. We show that this approach can help strengthen learning agents to successfully explore the new environment where long-term memory is essential. We have examined our approach in a challenging grid world environment and preliminary Gazebo experiments. Our experimental video is in the https URL.

"Google Deep Mind has created a neural network to learn how to play video games like human like a conventional Turing machine or access external memory of a neural network. The plan, AlphaGo Recently submitted a headline when it successfully beat Go Champion Fan Hui.This is the first time artificial intelligence hit a professional Go player - this is impossible for many people to predict I will

It developed a prominent position when DeepMind developed a neural network that learned to play video games by analyzing the behavior of pixels on the screen. In addition, Neurograph machine which can access external memory is constructed. In 2016, DeepMind participated in the use of NHS patient data to train the medical system at the Royal Free Hospital, and the Secretariat of the UK Information Committee decided that it violated the Data Protection Act. AlphaGo, developed by DeepMind researchers, won the first match with professional athletes in 2015, defeated World No. 2 player Lee Sedol in March 2016, and in 2017 the number one player Kegel Alpha Go I broke. The network trains by playing people and computers, using the Monte Carlo tree search algorithm to find the movement. Its success is very important as artificial intelligence researchers believe that the game of Go is a problem and do not expect humans to immediately lose to the computer.