“We can’t actually reproduce the environment of the moon moon the of environment the reproduce actually can’t “We There is a ten second delay in communications from the the from communications in delay second aten is There The company is a pioneer in affordable space robotics robotics space affordable in apioneer is company The Astrobotic’s Director of Guidance, Navigation, and and Navigation, Guidance, of Director Astrobotic’s As a Lunar XPRIZE team, Astrobotic’s core activities activities core Astrobotic’s team, XPRIZE aLunar As of aspinoff as 2008 in founded Technology, Astrobotic obstacle for the engineering team is being stuck on earth. on stuck being is team engineering the for obstacle biggest the –but surface lunar the on once operational can take pictures as it approaches the lunar surface surface lunar the approaches it as pictures take can cameras of anumber We have simulator. a3D use is do can cutting edge of the emerging commercial space industry. industry. space commercial emerging the of edge cutting XPRIZE WITH NVIDIA GPU NVIDIA WITH XPRIZE LUNAR GOOGLE’S EDGE FOR TECHNOLOGY GAINS ASTROBOTIC and use all of this information to properly position itself. itself. position properly to information this of all use and and data collection. Astrobotic is currently one of 23 23 of one currently is Astrobotic collection. data and autonomous landing, vehicle design, and simulations for for simulations and design, vehicle landing, autonomous governmental organizations with lunar payload delivery delivery payload lunar with organizations governmental Control. “On earth we have atmosphere, we have clouds, clouds, have we atmosphere, have we earth “On Control. Carnegie Mellon University’s Robotics Institute, is on the the on is Institute, Robotics University’s Mellon Carnegie motion over time in an optical flow. Our plan is to use all of all use to is plan Our flow. optical an in time over motion meet its demanding design and analysis needs. analysis and design demanding its meet robot travel 500 meters over the lunar surface while while surface lunar the over meters 500 travel robot revolve around the development of vision algorithms for for algorithms vision of development the around revolve sending video, images and data back to earth. Teams have Teams have earth. to back data and images video, sending that have and moon’s surface, the on arobot land safely NVIDIA customer, Astrobotic relies on GPU acceleration to to acceleration GPU on relies Astrobotic customer, NVIDIA until the end of 2015 to meet these goals. A longtime Alongtime goals. these meet to 2015 of end the until CHALLENGE both the launch and the landing, and still be fully fully be still and landing, the and launch the both we don’t have the same topography – so the best we we best the –so topography same the don’t have we Peterson, test,” Kevin landing explained physical any for full-time engineers is to design a vessel that can survive survive can that avessel design to is engineers full-time ten Astrobotic’s for task main The roving. and landing these data points as input for landing, where our lander lander our where landing, for input as points data these track and surface the map and moon the by fly that total of $30 million for the first privately funded team to team funded privately first the for million $30 of total –a XPRIZE Lunar coveted Google’s for competing teams and scientific, commercial, servicing technology, s “We often produce shell CAD models of our original designs designs original our of models CAD shell produce often “We The simplified models we used didn’t always correlate to correlate didn’talways used we models simplified The Astrobotic’s CIO and Director of Propulsion, needed a way away needed Propulsion, of Director and CIO Astrobotic’s challenge is simulating accurate launch A secondary The Astrobotic lunar rover, designed with NVIDIA GPUs NVIDIA with designed rover, lunar Astrobotic The of unique parts with more than 4,000 fasteners – and –and fasteners 4,000 than more with parts unique of explained Calaiaro. “But this simplification led to problems. problems. to led simplification this “But Calaiaro. explained time,” less takes testing so intense computationally all this data, the simulations are very computationally- very are simulations the data, this all with that is end our on challenge The autonomously. real-world test results, which meant that we had to go back back go to had we that meant which results, test real-world for simulation analysis because they’re less less they’re because analysis simulation for Calaiaro, Jason launch. during experience would aspacecraft vibration the to subjected when perform would lander, so it’s imperative that it can guide and land itself itself land and guide can it that it’s imperative so lander, intense and must be photo-realistic.” be must and intense team to run simulations using simplified shell CAD models. models. CAD shell simplified using simulations run to team his forced that bottlenecks computational overcome to element individual each how account into take must testing 100’s includes model spacecraft typical A place. first the in ground the off it makes vessel the that ensure to testing,

ASTROBOTIC CASE STUDY ASTROBOTIC CASE STUDY Tesla K20 boards – in order to achieve further speed speed further achieve to order –in boards K20 Tesla 1-2 weeks in design time alone,” he said. “On average, I see Isee alone,” time average, “On design in said. he 1-2 weeks As NVIDIA customers, Astrobotic’s engineers had already already had engineers Astrobotic’s customers, NVIDIA As of NVIDIA Corporation. All company and product names are trademarks or registered trademarks of the respective owners with which they they which with owners respective the of trademarks trademarks registered registered or and/or trademarks are are associated. trademarks names are Quadro product and NVIDIA and company All logo, NVIDIA the Corporation. NVIDIA, NVIDIA of reserved. rights All Corporation. NVIDIA © 2013 www.nvidia.com/quadro to go Quadro, NVIDIA about more To learn the new rover for wheels the of acomponent of image analysis Sample experienced significant performance benefits over the benefits performance significant experienced even though each test only took 5 or 10 minutes on one GPU, GPU, one on minutes 10 5or took only test each though even acceleration for photo rendering. realistic a lot of trial and error in trying to strike the right balance balance right the strike to trying in error and trial of a lot was There tests. more do and complexity the increase and approach design and design analysis.”approach –and GPU extra the with analysis in increase speed a 40% generation – including one Quadro K2000 board and two two and board K2000 Quadro one –including generation Using iray also increased Peterson’s lunar surface renders renders surface lunar Peterson’s increased also iray Using Calaiaro experienced significant performance increases in increases performance significant experienced Calaiaro design mechanical for acceleration GPU leveraging by CPU results, and that, compounded with the time spent creating creating spent time the with compounded that, and results, minute on a CPU. Peterson explains: “To put it in in it “To put explains: Peterson aCPU. on minute that and simultaneously, ANSYS and SolidWorks run about saves and workflow my streamlines which model, SOLUTION support, the second Tesla GPU would deliver an extra extra an deliver would GPU Tesla second the support, straight from design to analysis without having to bother bother to having without analysis to design from straight savings really add up. With this added GPU power I can also also Ican power GPU added this With up. add really savings Peterson and Calaiaro saw immediate performance performance immediate saw Calaiaro and Peterson GPUs NVIDIA their upgrade to opted Calaiaro and Peterson use a solid mesh calculated directly from my original CAD CAD original my from directly calculated mesh asolid use boost. They also switched to NVIDIA iray to leverage GPU GPU leverage to iray NVIDIA to switched also They boost. both SolidWorks and ANSYS. “Now I can easily just go go just easily Ican “Now ANSYS. and SolidWorks both the of quality the and complexity of level the between bi-directionality is incredibly powerful and changes how we we how changes and powerful incredibly is bi-directionality from the older Fermi generation to the current Kepler Kepler current the to generation Fermi older the from with building simplified shell CAD models; I can instead instead can I models; CAD shell simplified building with when you have to test 100 different variables, those time time those variables, different 100 test to have you when improvements with Kepler-based Quadro and Tesla GPUs. GPUs. Tesla and Quadro Kepler-based with improvements multi-GPU providing now 14.5 ANSYS with And increases. ANSYS. in analysis element finite and SolidWorks in the shell models, was impacting our tight schedule.” tight our impacting was models, shell the to 20 frames per second on a GPU, up from 1 frame per per 1frame from up aGPU, on second per frames 20 to “For us, everything needs to be done yesterday,” done be to added needs everything us, “For amore make to you allows it because key is “Analysis “The bottom line is that we’re a small company trying to to trying company we’re asmall that is line bottom “The Astrobotic’s engineers to handle intricate designs designs intricate handle to engineers Astrobotic’s our counterparts. our that times than is aboutcycle faster 4-6 GPUs allow usto haveNVIDIA adesign Applications Graphics Workstation Used: &Software Hardware of two GPUs, I’m able to test a far more accurate model and and model accurate more afar test to I’m able GPUs, two of or government agencies would probably spend more than than more spend probably would agencies government or corners.” counterparts – for instance, we will spend 6 months 6months spend will we instance, –for counterparts a have to us allow GPUs NVIDIA possible. as efficiently allow us to keep a competitive edge without cutting cutting without edge acompetitive keep to us allow leads which compromises, no with analysis immense and and NVIDIA’s Kepler GPUs are enabling the power and and power the enabling are GPUs Kepler NVIDIA’s and We simulations. on rely we so expensive, prohibitively are get the results I need faster.” Ineed results the get do things on a tight schedule with limited resources,” resources,” limited with schedule atight on things do our than faster times 4-6 about is that cycle design making decisions on our feet and moving forward as as forward moving and feet our on decisions making possible landing scenario; they need to be very high-fidelity, high-fidelity, very be to need they scenario; landing possible simulations surface lunar different amillion run to need tests Flight It’s huge. minutes. ten in render same that run and render the up set to have would we before perspective, In the race to the moon, speed and accuracy are key. are accuracy and speed moon, the to race the In speed that we require.” Peterson concluded. “NVIDIA’s GPUs and iray renderer renderer iray and GPUs “NVIDIA’s concluded. Peterson There quickly. it do but right, it do “Always Peterson. allow GPUs Tesla and Quadro Kepler-based NVIDIA’s IMPACT building our rover, while comparable industry members members industry comparable rover, while our building before launch to make sure that we’ve tested every every tested we’ve that sure make to launch before very accurate, but we also need the ability to iterate quickly, quickly, iterate to ability the need also we but accurate, very and down sit Ican Now overnight. run to it leave literally informed design,” said Calairo. “Now with the capabilities capabilities the with “Now Calairo. said design,” informed isn’t a lot of time to step back and take pause; we’re always we’re always pause; take and back step to time of isn’t alot to more accurate results. accurate more to two years on a similar effort.” asimilar on years two Solidworks, ANSYS Solidworks, K20 Tesla +2x K2000 Quadro Z800 HP