Скачать с ютуб 4 Hardest Things to Teach a Robot | What the Stuff?! в хорошем качестве

4 Hardest Things to Teach a Robot | What the Stuff?!

Being a human isn’t always easy, but it’s way less complicated than building a human! Article: 10 Hardest Things to Teach a Robot http://science.howstuffworks.com/10-h... Music Attribution: "At the Count" by Broke For Free Video Attribution: Robocup    • Robocup - The Small Sized League   RobotEye Viewer    • RobotEye Viewer - PCL Ocular Robotics...   DLR Robotic Motion Simulator    • DLR Robotic Motion Simulator - Drivin...   Bimanual Dexterous Manipulation for Autonomous Service Robots    • Bimanual Dexterous Manipulation for A...   Beyonce Live at NFL Superbowl 2013    • Видео   What the Stuff?! episodes are available every Monday and Friday at Noon ET. Subscribe http://bit.ly/1AWgeM7 Twitter   / howstuffworks   Facebook   / howstuffworks   Google+ https://plus.google.com/+howstuffworks Website http://www.howstuffworks.com Watch More    / howstuffworks   Being a human isn’t always easy, but it’s way less complicated than building a human. Lots of things that are simple for people are really hard to teach robots: Trailblazing: Humans can navigate a room or a city pretty easily. But when you break that task down – you’ve gotta perceive and make sense of your environment, form a plan, put yourself into motion, and account for unforeseen obstacles – you see how it’s tricky to program into a robot. Stereoscopic cameras or laser scanners can act as eyes, but the bigger challenge is getting the ‘bot to make good decisions via SLAM: simultaneous location and mapping. Which means continuously creating and updating a map that includes its own position. Dexterity: Getting a robot to move around without crashing into stuff is hard. But it’s even harder to have a robot to interact purposefully with specific stuff. Humans are dexterous: Our vision, skin, nerves, and musculoskeletal systems let us immediately determine how to reach a given object and pick it up based on its shape and squishitude. Here, hardware is key. To interact with varying objects, robots need physical flexibility, aka compliance. Things like spring joints, pressure-sensitive contact surfaces, infrared sensors, and pliable skin. Human Speech: Just understanding us, let alone holding a conversation, is a challenge in robotics. No two speakers use the exact same phrasing, vocab, or dialect. A single idea may be expressed thousands of ways, so it’s proven mildly impossible to create rules that let robots process natural language – that’s normal human speech. Statistical analysis helps – you can feed a robot a massive collection of text (aka a corpus) and have it analyze what words often come together and in what order. Thus the ‘bot learns context: Like bay versus Bey. Deception: The fine art of deception helps animals best competitors and predators – and that goes for humans, too. Like in sports or high-stakes business. But deception requires imagination. Robots are aces at processing input and creating statistical models, but not so great at forming ideas outside of that existing data. But a team at Georgia Tech is working on it – with help from squirrels. Squirrels protect their food by leading competitors to old, empty caches. The team coded those behaviors into simple rules (aka algorithms), which their robots could use to determine when to send out false data to lead another ‘bot astray. Now, whether you want robots to know how to be deceptive – that’s another question entirely. Here’s a question for you: What do you wanna teach robots to do? And hey, if you liked this video, make it official - and subscribe so you won’t miss the next one. To learn more, like how future robots might act ethically and parse emotions, check out our article The 10 Hardest Things to Teach a Robot on HowStuffWorks.com.