Crash testing with these massive mammals has come a long way from using real cadavers. More information Adding us as a Preferred Source in Google by using this link indicates that you would like to see more of our content in Google News results. The moose crash test dummy helping Volvo engineers in Sweden build cars. Breakthroughs, discoveries, and DIY tips sent six days a week. By signing up, you confirm you are 16+, will receive newsletters and promotional content and agree to our Terms of Use and acknowledge the data practices in our Privacy Policy .

State-of-the-art reinforcement learning algorithms mostly rely on being allowed to directly interact with their environment to collect millions of observations. This makes it hard to transfer their success to industrial control problems, where simulations are often very costly or do not exist, and exploring in the real environment can potentially lead to catastrophic events. Recently developed, model-free, offline algorithms, can learn from a single dataset by mitigating extrapolation error in value functions. However, the robustness of the training process is still comparatively low, a problem known from methods using value functions. To improve robustness and stability of the learning process, we use dynamics models to assess policy performance instead of value functions, resulting in MOOSE (MOdel-based Offline policy Search with Ensembles), an algorithm which ensures low model bias by keeping the policy within the support of the data. We compare MOOSE with state-of-the-art model-free, offline RL algorithms BEAR and BCQ on the Industrial Benchmark and Mujoco continuous control tasks in terms of robust performance, and find that MOOSE outperforms its model-free counterparts in almost all considered cases, often even by far.

Video Friday is your weekly selection of awesome robotics videos, collected by your Automaton bloggers. We'll also be posting a weekly calendar of upcoming robotics events for the next few months; here's what we have so far (send us your events!): Let us know if you have suggestions for next week, and enjoy today's videos. NASA is sending a small helicopter to Mars in 2020, and it managed to get airborne in a simulated Martian atmosphere without crashing or exploding. I really want to get excited about this thing, and from a technology perspective, I am.

Data preprocessing is the first (and arguably most important) step toward building a working machine learning model. If your data hasn't been cleaned and preprocessed, your model does not work. Data preprocessing is generally thought of as the boring part. But it's the difference between being prepared and being completely unprepared. You might not like the preparation part, but tightening down the details in advance can save you from one nightmare of a trip.

As social media shrinks and quickens the world, and the threats we face seem to grow ever taller and closer, the relevance of entertainment (and, whisper it, art) appears to diminish. Surely the king's ransom it takes to fund a blockbuster film would be more usefully and perhaps profitably applied to combating climate change, or Boris Johnson's gaffes? What is a commissioned oil portrait if not the most extravagant of all selfies, taken in a world to whose indignities and injustices no one can claim to be blind? In this context, video games can seem like the hollowest endeavours of all. They cost ever larger sums to build, require ever greater numbers of talented minds to make, and distract ever more humans from the practical issues of reality.

I remember my roommate, way back in 1986, laboriously stringing together phonemes with Apple's Macintalk software to get his Mac to utter a few sentences. It was pioneering at the time -- anybody else remember the Talking Moose's jokes? But boy, have things improved since then. Publishing a new round of papers on its new machine learning journal, Apple showed off how its AI technology has improved the voice of its Siri digital assistant. To hear how the voice has improved from iOS 9 to iOS 10 to the forthcoming iOS 11, check the samples at the end of the paper.

How do you train a car to drive itself? Let it watch real drivers. Engineers from graphics processing unit (GPU) company Nvidia designed a system that learned how to drive after watching humans drive for a total of 72 hours, as reported by NetworkWorld. The likely conclusion of the system's success is that driverless cars are coming faster than most of us expected. The details of how Nvidia trained two test cars are in a file titled End to End Learning for Self-Driving Cars.