ZeroBreeze Latency Tracking

Why do I hear latency when using the Direct Monitoring feature on my interface? & Focusrite Audio Engineering
Applies to: Scarlett (whole range), Saffire (whole range)
With Direct Monitor engaged (Scarlett Solo, Scarlett 2i2, Scarlett 2i4),
“Zero Latency Tracking” preset selected within Scarlett MixControl (other Scarlett interfaces) or within Saffire MixControl (Saffire interfaces), the signals connected directly to the interface should pass through directly to monitors or headphones without noticeable latency.Typically your recording software has the ability to monitor the signal as it comes through it, and if direct monitoring is being used, software monitoring within your DAW should be disabled to avoid hearing the signal twice – once from the direct monitoring and again from the output from the recording software.
This is because you hear the direct sound (before it is passed to the computer) and then you hear the same audio that has been processed by the computer afterwards. Muting the track you are recording into prevents this, as you will only hear the recording source directly.
For more information on latency issues with interfaces, click .
If you are hearing the direct signal but are getting another signal with latency without your recording software opened, there could be a setting that needs to be disabled within the Sound Properties of the computer (this applies to Windows only).To get to the Sound Properties, go to Windows & Control Panel & Hardware and Sound & Sound. This will open the Sound Properties where the device would be selected as the default device for Playback and Recording.Under the Recording tab, select the interface and click Properties.
When the Line In Properties window opens, select the Listen tab. From here, make sure the “Listen to this device” option is not selected. If this option is selected, it will allow the direct signal to pass through the Windows driver and could result in latency on the output of the Scarlett and Saffire interface.
Have more questions?
Didn't find what you were looking for?
Search again using our search tool.Zero Latency: The VR revolution begins in Melbourne, Australia - CNET
CNET también está disponible en espa?ol.
Don't show this again
Check out all the places we've been on CNET's Road Trip 2015.Stepping onto the large freight elevator, my body sways to keep balanced as the rising platform lifts me into the zombie-infested city above. Yet my conscious mind knows I'm actually standing on unmoving concrete and that the sense of motion is all in my head.I've been playing Zero Latency for mere moments and already I'm amazed at how it overwhelms my senses. This is the world's largest virtual reality (VR) attraction, and its doors open August 15 in North Melbourne, Australia.Once confined to the realm of science-fiction movies like Walt
Disney's "Tron," virtual reality has grown into a real-world industry . But since the Oculus Rift virtual reality headset appeared on Kickstarter in 2012 and reignited passions for VR, the big question has been when these prototypes and concepts will launch for public consumption. And when we reach commercial launch, will the experience live up to the hype? Zero Latency is one of the first ventures that will put mainstream viability to the test.
Scott Vandonkelaar had the seed of the idea behind
in 2012, months before the Oculus Rift's record-breaking Kickstarter campaign had shipped its first prototype virtual-reality headset, and years before game developer Valve and smartphone maker HTC partnered to announce the HTC Vive VR kit.
I've been playing Zero Latency for mere moments, and already I'm amazed at how it overwhelms my senses.
Vandonkelaar had been working on an optical tracking system to track battles of radio-controlled battleships armed with BB-guns, but when he saw the success of the Oculus Kickstarter campaign, he realized his tracking system could work beautifully when combined with VR.Like the HTC Vive, Vandonkelaar's idea allowed players to walk around a real-world, physical space whilst wearing a head-mounted display. However, Vandonkelaar had grander ambitions about the size of the playing area. Instead of keeping the action confined to a single small room like Vive, he wanted to place up to six players in a much larger space.Vandonkelaar raised the idea with a few ex-workmates. "I started mentioning that I had this idea -- do I sound like a crazy person?" Zero Latency soon formed and grew to a team of six full-time staff with extra contractors brought on board for specialist duties.
Want to play Zombie Fort: Smackdown?Building the technology has taken the team at Zero Latency more than three years. While initial funding came out of the team's own pockets, the first round of public funding came courtesy of crowdfunding platform Pozible.com in July 2014, eighteen months after the idea's conception. Tim Ruse, director and resident business guy at Zero Latency said that "it was the hardest thirty grand we've ever earned."Next stop on the funding tour was Film Victoria, a state government funding body, where a grant for $60,000 was approved. At the time the team hadn't come up with a name for the game, so they lodged the grant application under the title "Zombie Fort: Smackdown."The Zero Latency warehouse floor.
Bennet Ring/CNET
The team first built working prototypes of the technology mounted inside primitive plywood cases, but they were functional nonetheless. Ruse credits this prototype-led approach with the next milestone in the money-raising process. It was the helping hand of Dean Dorrell, from boutique investment firm Carthona Capital, that gave the company a $1 million cash injection."He sent me a Google Hangout message, of all things,"Ruse said of Dorrell's unorthodox approach. "He said he was really interested in investing, and he came down on the Thursday.""He said, 'I really like it, I reckon you need a million bucks,'" said Ruse. "I was like, 'Alright, off you go,' not expecting to hear from him again. Then he was down with another business partner and he looked serious. It all fell into place from there."After three years of research and development, Zero Latency is ready to open its doors to the public on August 15. The game offers a one-hour cooperative experience for four or six players, and tickets are currently selling for AU$88 (about ?40 or $65) per person. Online sales began in early August, and in a week, hundreds of eager gamers had already signed up for this world-first experience.Suiting upAs we enter the large warehouse space, I'm amazed by just how big the area is. All 400 square metres (4,300 square feet) are covered in a white grid pattern to aid camera rigs mounted above the playing area in monitoring each player's movements. Covering the entire play space are 129 Sony PlayStation Eye cameras driven by the team's proprietary software. They're arranged in circular rings that cover a 360-degree view, each ring with its own dedicated controller PC.According to Vandonkelaar, the team tried to use off-the-shelf hardware where possible, but the immaturity of VR made this difficult."VR is all really new, so we're working with companies that still have all of their equipment in the prototype stage. So nothing is locked down, and a lot of the features that are meant to work don't."The players are tracked by 129 PlayStation Eye cameras.
Bennet Ring/CNET
As the games master suits us up for the battle ahead, I'm pleasantly surprised by the weight and comfort of the backpack. Each pack houses a heavily customised Alienware Alpha PC to render each player's view of the world. Looking like a Proton Pack from "Ghostbusters," it sits easily in the square of my back, despite including its own power source. The Oculus Rift DK2 and a set of quality headphones are tethered to the backpack, and a glowing orb sits on top of the headset.We're then handed our weapon for the impending zombie apocalypse: a large five-pound rifle machined from a solid block of PVC plastic on the massive computer-controlled milling machine, tucked away in a workshop at the back of the warehouse.A button at the base of the magazine swaps between machine gun, shotgun and sniper modes, while another button at the front of the magazine is used to reload. Finally, a pump action at the front is used to reload special grenade ammo, as well as shotgun and sniper rounds.Clear and present greatnessAs I slip the Rift DK2 over my eyes, the cold, empty, concrete warehouse around me is replaced by a military firing range that wouldn't look out of place in "Aliens." At night, the warehouse drops into complete darkness when the game begins. You're running blind in a cavernous space, the virtual world your only guide to not slamming into a wall in the real one.While the game was built using a relatively simple modern game engine called Unity, I'm impressed by just how real the world around me feels. Commonly referred to as "presence" by VR pundits, this sensation of being in a real space intensifies when I walk forward to the firing range, before raising my weapon to let off a few virtual rounds.The VR headset and custom Alienware Alpha computer backpack that make Zero Latency possible.
Bennet Ring/CNET
This experience is massively different compared with the many hours I've spent strapped stationary into my home racing cockpit with my own Rift DK2 in place. Despite the lack of physical haptic feedback, firing the weapon still feels meaty and impactful, though the accuracy isn't quite good enough to aim down the sights. Besides, it's surprising how heavy the weapon becomes after just five minutes of zombie-blasting. Firing from the hip is entirely feasible thanks to the gun's virtual laser sight.The next 40 minutes are spent walking through a variety of virtual locales, from open city streets to cramped laboratories. The warehouse itself was around 100 metres (328 feet) in length, but clever use of loading screens and level design mean I ended up walking over 10 times that during my experience. Combine this with the weight of the gun and by the end of the hour I was exhausted, dripping in sweat and my arms ached.When my headset is sitting just right on my face, I am completely in the moment, ignorant of the fact that I'm actually walking around an empty warehouse. In my mind I'm gunning down zombies and rebels by the dozen, with my real buddies covering my virtual flanks. Yet some of the best moments are when there are no enemies at all, where the lack of immediate threat means I can relax and simply enjoy exploring the virtual environments around me.Step up to the virtual firing range.
Zero Latency
Day one of the VR revolutionIt's the best virtual reality experience I have ever had. And I have tried them all since Dactyl Nightmare almost made me puke back in 1991. It's not entirely perfect though -- the Rift DK2's mounting straps aren't ideal for such a physical VR experience, and I had to readjust the headset every few minutes to line up the viewing sweet-spot. Hopefully, the team at Zero Latency redesigns the mounting system entirely, so that the headset doesn't move, no matter how physical each player gets.
Zero Latency and, by extension, VR totally transcend what is meant by immersive gaming.
There are also a couple of buggy spots where the tracking wigs out, instantly confusing my sense of balance in a way that I might associate with a few too many beers. But overall, these are minor inconsistencies in an overwhelmingly amazing gaming experience. Zero Latency and, by extension, VR totally transcend what is meant by immersive gaming. Considering how powerful this early experience is, it's no surprise to hear that the team at Zero Latency is already in discussions with other partners to export the technology around the globe.According to Ruse, "The version you play today is commercial version one, in terms of what we are going to box and ship early next year to different venues. We've got people overseas who are interested, all over the place. It's just sorting out who will be the best partners."
The team is also considering changing the game experience from a purely cooperative game to one that involves adversarial combat. "We built this elaborate and really good fun co-op mission," Ruse explains, "but sometimes the best 10 minutes people have is shooting their mates at the end."If this is day one of the VR revolution, I can't wait to see where we are in a few years.
Road Trip 2015
Zero Latency: The VR revolution begins in Melbourne, Australia第34期:跟Hoevo学习VR/AR/MR的空间定位技术 | 硬创公开课 - 干货必读 - 优派YOPAI.COM
o 1年前发布
编者按:今年 8 月,雷锋网(搜索“雷锋网(搜索“雷锋网”公众号关注)”公众号关注)将在深圳举办一场盛况空前,且有全球影响力的人工智能与机器人创新大会。届时雷锋网将发布「人工智能&机器人 Top25 创新企业榜」榜单。目前,我们正在拜访人工智能、机器人领域的相关公司,从中筛选最终入选榜单的公司名单。如果你也想加入我们的榜单之中,请联系:。嘉宾简介:缪庆,Hoevo公司CEO。曾计算机竞赛一等奖保送同济大学建筑本科,后在英国建筑学名校Cardiff Uni攻读PhD(Genetic Algorithm in sustainable design optimisation)主攻基因算法优化设计问题,同导师Andrew Marsh开发建筑节能设计软件Ecotect,06年出售与Autodesk。软件出售后辍学PhD,终获哲学硕士。在伦敦Gensler担任多年建筑师,参与设计了迪拜金融中心,上海中心等项目。随后移居珠海,从零创办企业为壳牌石油生产Helix机油。2014年售出工厂股权回到上海重新创业,创立了Hoevo。本文整理自缪庆在雷锋网硬创公开课的分享,以下是全文内容。▌VR、AR、MR的区别是什么?VR(Virtual Reality)VR是基础概念,但是要确切定义又很难。一般公认的是VR一定是沉浸式的,用技术手段向使用者复制了在某个现场的体验。这种体验可以是一个真实的现场(比如用360°摄像机拍摄的,我们可以称为360°视频),也可以是一个虚拟的现场(比如用Unreal/Unity建造的虚拟内容)。沉浸式定义并不清晰,是有两个级别标准的:第一级3DOF(Degrees of Freedom,通常缩写为DOF)沉浸式体验只包括360°的视野,你可以自由旋转身体或者头部来观察。比如GearVR为首的插手机的盒子VR都是这类。因为360°视频即使在可见的未来依然是3DOF的,因此视频这一类VR的主要概念,而且这一类VR因为Google Cardboard普及程度最高。第二级6DOF不仅可以旋转还可以在这个现场里自由移动,Oculus、HTC Vive、PSVR(PlayStation VR)都是这一类。因为不论真实世界,还是虚拟内容/虚拟世界本质是6DOF的,具备6DOF的设备才能在体验上无限逼近真实而产生身临其境的错觉,才能对于游戏等虚拟内容有完整的体验。而目前所有设备能提供的6DOF的范围是非常有限的。(顺便说一句,9DOF/9轴是传感器概念,计算结果依然是6DOF的。)AR(Augmented Reality)AR是最容易解释的,但是却常常和MR混淆,因为两者都涉及到虚拟内容和现实内容的结合。如果用我们的定义,AR是在现实世界/现实视频上叠加了一层虚拟的信息,这些虚拟信息本身和世界的位置/角度无关,你不会误会它们是世界的一部分。就像有些汽车的挡风玻璃上可以显示车速,Google Glass就是这一类应用。MR(Mixed Reality)MR最直接的例子,就是Magic Leap的视频那条在体育馆里跃起的鲸鱼。MR技术将虚拟内容和现实叠加后呈现在你的眼前。MR的必要条件是6DOF,因为必须根据你的走动和视角即时的计算出虚拟内容的大小、角度、光影。技术的发展会使你越来越无法区分虚拟内容和现实。从技术上MR比VR要求的定位精度和响应速度更高,技术上是一个更大的挑战。有趣的是难易会互换,在优秀的技术支持下,从内容开发者角度来说MR的开发量比VR小的多。举个例子,如果我们想在一个精美的酒店的桌子上摆一个虚拟的花瓶。用MR技术只需要制作虚拟花瓶即可,而用VR则必须把精美的酒店和桌子这些现实物体重建一遍。换句话说,MR可以享用虚拟内容的利益而避开包括现实重建、感官丧失等成本。这也许可以解释为什么Magic Leap比Oculus估值高那么多。从本质上说,VR仅仅是100%覆盖虚拟内容的MR而已,VR设备可能是MR设备过渡阶段,因为MR设备都可以实现VR。▌VR的定位技术标准是什么?Oculus的CEO Brendan Iribe在2014年Oculus Connect讲到VR沉浸的5个要点,我们借其中的1、2两点来试着统一一下什么是VR/MR的定位必要标准:1、追踪(Tracking)6DOF(不仅能够追踪旋转,也能追踪移动),360度追踪(不管朝向哪个方向都能追踪到6DOF);毫米级或更高精确度,无颤抖噪音(被追踪物体静止时不会出现抖动);舒适的追踪范围(头盔所处的空间定位面积)。2、延迟(Latency)从动作发生到屏幕显示最后一帧画面的时间低于20毫秒;整合光学追踪和IMU数据;循环最小化: tracker → CPU → GPU → display → photons在与定位相关的这两点里,我们可以总结以下几条标准:1、6DOF(可以定位旋转和移动),360°;2、毫米级或更高的精度,没有抖动噪音(完全静止时定位不会抖动);3、舒服的定位范围(这里Oculus没有提出舒服的范围是多大,不免有为自己技术辩护之嫌);4、从运动发生到显示在屏幕上的时间差&20ms,即,延迟&20ms,刷新率&50Hz。我们在这里加上重要的一个条件:必须满足多人同时使用。▌目前主要有哪些空间定位的技术?空间定位是一个巨大的话题,我想把问题缩小一下有哪些潜在能为VR/MR服务的室内定位技术,这里就不用讨论GPS这类大范围定位了。定位数据的描述是我们讨论的基础。在此我尽量避免太复杂的数学,但是因为各种传感器技术获得的数据即使都是关于位置的信息,有的是加速度有的是速度,为了帮助理解,我简单重复一下中学物理和数学里的概念。定位是要知道一个物体瞬间(时刻k)的位置和角度。以位置为例,可以用来描述。第一阶是位置,第二阶是速度,第三阶是加速度。最理想的是直接知道位置,但是知道速度也有帮助,用 x1=x0+v*dt 可以通过上一时刻的位置和速度、时间求得此刻的位置,可以对位置数据进行矫正;同理三阶的加速度v1=v0+a*dt可以用来矫正速度。角度同样有三阶,角度,角速度,角加速度,我接触的传感器和算法之内没有用到角加速度的,所以旋转可以仅用两阶。基本逻辑是,获得越低阶的数据越好,用上一阶的数据可以给低一阶提供矫正。也可以跨两阶来矫正,当然必须建立更复杂的数学模型对中间阶进行描述。但是可不可以用高阶观察值直接计算低阶数据? 答案是明确的不可以,IMU不能定位位置的原因也正在这里,后面会解释。Breandan提到将光学定位和IMU定位结合,我们从IMU讲起:1、&IMUIMU尤其是因为InvenSense和Bosch两家的努力,MEMS级别的IMU成本很低而且非常成熟,每个手机里都有一组(iPhone 有两组),它使得Gear VR和同类VR能3DOF定位。下面以Oculus的IMU设计为例:第一部分是Bosch Bmi055 由一个加速度计一个陀螺仪组成。加速度计的值描述了加速度的方向和大小,因为重力的存在而且重力方向唯一,我们可以计算出三阶的位置信息(x,y,z方向的加速度);和一阶的旋转信息的两根轴(x,y),垂直重力方向的平面内的旋转依然是未知的。而陀螺仪的值描述了二阶旋转信息,三轴的旋转的速度,如前面解释的,可以用来协助矫正加速度计提供的一阶旋转信息。第二部分是Honeywell HMC5983(疑似),这颗芯片类似于一个指南针,可以提供磁力的方向和大小。因为地球磁场总是和地心引力垂直,它正好提供了加速度计里我们缺失的一阶z的旋转信息。至此我们获得完整的一阶旋转数据和二阶旋转数据。但是关于位置信息,我们只有3阶的加速度数据,无法提供有效的位置信息。顺便提一下,iPhone同时配有InvenSene MP67B 和Bosch BMA280,二颗加速度计。我们认为是用来矫正噪音。因此,用两颗IMU来矫正数据并不完全是异想天开,虽然其它没有一家用这样的方案,但iPhone都这么干了。HTC&Vive的内置IMU也是基于类似的配置。虽然IMU可以提供完整的一阶和二阶角度数据因此可以用来计算空间角度,但是磁场并不可靠,靠近很多电器或者金属磁场都会改变,因此会产生drift的现象(即在z轴平面内错误旋转)。IMU不能计算位置可以看和Tango 老大Johnny齐名的UC Davis大神Oliver的视频,希望他可以说服你。用加速度数据来计算位置是一个很tricky的事情。必须符合两个条件才可以计算:1、必须能不断精确矫正初速度,2、两次矫正时间间隔越久精度越低。我们走路的脚的移动数据是极少数符合的应用,落地时速度为0 m/s可以用来矫正,下一次落地间隔不太久。放IMU在两只鞋子里,就能用三阶模型计算出接近的跑步距离。但是无论如何,IMU只能提供精确的角度数据,不能提供满足VR使用标准的位置数据。IMU的刷新率很高,Vive IMU的刷新率在1006Hz(很奇怪的数字)。某宝上出售的IMU都有100Hz以上两倍于我们50Hz的标准。而且IMU的数据几乎不需要计算,有很多种ARHS的算法可以把传感器数据转换成3DOF数据。总结:6DOF:不能,只能提供3DOF信息。可提供三阶位置信息。单个成本:极低(10元左右)计算代价:极低(单片机都可以)大面积覆盖成本:零这里插入讲一个异类STEM。基本逻辑是用人工建立强大的磁场。因为使用环境内只能有一个磁场,而磁场衰减非常快,因此我没有深入研究,推测是用多个磁力计的方向三角计算获得6DOF。感兴趣的同学可以研究了探讨一下。2、&光学定位(PSVR、Oculus、Optitrack、TheVoid、ZeroLatency)Daniel DeMenthon和Larry S. Davis在1995年在International Journal of Computer Vision发了一篇论文, 'Model-Based Object Pose in 25 Lines of Code'。原文下载这篇论文建立了一套我们称为POSIT的方法,这套方法建立了整套光学定位的基础。POSIT是通过透视结果(近大远小)计算出物体相对于光线采集设备(比如CMOS)的旋转和位移。没错, POSIT能直接计算出完整的一阶位置和旋转数据。但是,光学定位都需要外部相机。Oculus配备了Constellation,PSVR 比较坏没有包含,但是必须有PS Eye才可以使用PSVR,也是个相机。比如Oculus,用的是红外LED发光,用没有红外滤镜的摄像头看就是这个样子。取得这张图片,然后用blob detect找到所有点的位置,然后用POSIT方法用点在图片中的x,y位置和原始的三维的位置x,y,z(设计时已知了)就可以计算出Oculus的6DOF。前面提到Oculus同时有IMU提供的3阶位置和2阶及1阶角度数据和光学定位计算出来的1阶位置和角度数据,每个数据有自己的噪音,需要一个方法来计算出一组唯一的数据。我们比较确定的是,Oculus用了一种R.E.Kalman在1960年的论文“A New Approach to Linear Filtering and Prediction Problems”提出的算法来融合和计算出唯一的6DOF数据。值得一提的是,因为获得数据(图像和IMU采集)的时间T0,计算完成时间T1,和最后画面投在显示屏上的时间T2之间有差(前面提到T2-T0&25ms)。所以这一套Kalman模型并不是直接给出T0信息的计算结果,而是通过2阶数据的计算预测出T2时的数据。这才是真正给出的6DOF,这就是世界最高标准定位算法的追求。&Oliver的测试照片测试发现,Oculus通过控制Led的发光频率来区分每一个点,这既有利于POSIT算法,也可以帮助系统区分两台Oculus,满足多人的要求。PSVR延续了PS Eye的方法,也是基于POSIT的光学定位。和Oculus的区别是,Sony一贯采用可见光。如图中的手柄,Sony用可见光的颜色来玩家和设备。还有一个差别是,以前Hack PSEye和PS Move时发现,PS的光学只采集1阶的位置信息,无法采集1阶的角度信息,角度信息完全由IMU提供,这点和Oculus不同。PSVR或者Oculus的摄像头可以通过阵列覆盖更大空间,这种方案最有名的公司是一家原本非常专业的动作捕捉MOCAP公司Optitrack。TheVoid采用了Optitrack来建立一套可以大范围行走的VR空间。Optitrack本质上是多个Oculus Constellation阵列来捕捉Marker,然后用POSIT方法来计算6DOF。而澳洲的Zero Latency则是用了多个PS Eye的阵列,算法和优缺点都是很接近的。上图国内一家VR公司的图,设备上的白色小球就是Optitrack的Marker。下图是Optitrack的相机。通过一圈红外发光来使白色的Marker反光,之后就像Oculus那样计算。Optitrack本来是用来拍电影或者采集动作的,因此原本就是为了大范围使用准备的。使用时,相机阵列如上图挂在上方。这种方法的好处是多个相机不容易遮挡,而且同一个Marker有多个相机捕捉到,因此可以通过fusion算法去除掉噪音(jitter)。光学定位就是对计算机成像(pin hole model)的逆运算,如果熟悉线性代数会很容易理解计算过程。浸润久了,你看什么都是一个矩阵了。这里不涉及复杂的原理,直接说几个结论,如果需要论证可以来探讨。因为光学采集是FOV固定,CMOS分辨率固定,因此捕捉分辨率和距离成反比。假设水平FOV 51度,我们用两种CMOS 640x480和,在1m,5m,10m的距离上每个像素代表的距离如下表:中间单位是mm。10m距离,如果用640x480来采集,发生14mm位移依然在一个像素内,意味着完全无法捕捉这个位移。以Brendan所说的mm级(sub-mm)精度为标准,我们需要很多很高分辨率的相机了。光学定位的POSIT算法计算成本不低。而且,每一个相机都需要用高速网线通信,通过Hub连到计算,然后分别计算再同步。虽然单个相机都可以通过独立线程计算,充分利用CUDA等并行计算的优势。随着相机数量的上升,计算和通信成本依然是指数级上升。因此虽然光学定位理论上可以大范围复制,但是实际上很快会遇到天花板。总结:6DOF:可以,一般辅以IMU 3DOF信息计算。单个成本:不低(高分辨率,高刷新率)计算代价:高(CPU配合GPU计算)大面积覆盖成本:与覆盖面积成指数级,成本迅速上升。起步价:5000元/m23、Lighthouse(HTC Vive)Valve这是VR世界的一朵美丽奇葩,创新了一种定位技术Lighthouse,同时创新了一种硬件合作方式,由Valve提供Lighthouse 技术HTC来生产HMD,后来Acer也模仿了这个模式。最初,Valve团队可能是打算研发光学定位(下图),某次采访中提到是太丑而放弃了,的确谁也受不了这样的房间。Lighthouse是一种非常聪明的技术,当然也不是Valve发明。最早在几个美国大学的实验和paper上有试验过类似的方法。Lighthouse 有两个,一般对角放置。两个之间有一个是Master,两者通过光学同步,同步之后Slave才会开始工作。每个Lighthouse里面有两个3600rpm Nidec 电机,就是硬盘里用那种,两个垂直放置(x,y),分别发出线性红外激光扫过整个空间。测试发现,具体工作时间是这样的:Lighthouse A&闪光Lighthouse A – X激光开,扫描房间。&Y激光关&(8.333ms)Lighthouse A – X激光关。Y激光开,扫描房间&(8.333ms)Lighthouse B 闪光Lighthouse B&– X激光开,扫描房间。&Y激光关&(8.333ms)Lighthouse B – X激光关。Y激光开,扫描房间&(8.333ms)不断重复。Vive拆开会看到32个红外二极管,当收到闪光或者激光扫描时二极管电压上升。如果我们取三个二极管E,F,G在X平面里看:Lighthouse A&闪光E,F,G都高电压,此时记为时刻T0Lighthouse A – X激光开,扫描房间。&Y激光关&(8.333ms)E,F,G被扫描到并获得高电压时刻依次为T1,T2,T3因为我们知道激光转速3600rpm,所以角速度是21.6 degree/ms。于是我们计算时间差dT,dT*21.6可以求得E,F,G两两之间和他们与LighthouseA 初始角度的夹角。因为E,F,G两两之间的距离已知,可以解得X平面内唯一的位置。(事实没有那么简单,因为EFG与X平面有夹角,但是依然可解)通过简单的数学,我们就可以计算出6DOF的一阶数据,非常聪明的方案。计算成本很低,设备和Lighthouse之间的距离远近不影响精度,这个相对光学定位来说是巨大的优势。Lighthouse技术的主要问题是8.333ms的扫描周期是排他的,只可以有唯一的光源进行扫描。如果加一个Lighthouse那就必须在时间上增加一个8.33*2的时间周期。所以大面积很难做到。也许可以通过7200rpm的电机降低单次扫描时间到4.1666ms,如果我们用四组lighthouse,就需要4.16*2*4一共33ms的周期,这就已经接近30hz了,如果用8组就是15hz,即使插值IMU数据也很容易产生晕眩了。有没有技术方案在Lighthouse的基础上扩大范围,我们团队对这个问题非常感兴趣。至少目前,我们无法在理论上解决这个问题。理论上,虽然可以通过与Lighthouse增加通信来控制基站开关,让一个设备大范围使用,但是多个设备大范围使用Lighthouse定位似乎是无解的。总结:6DOF:可以,一般辅以IMU 3DOF信息计算。单个成本:不低(高分辨率,高刷新率)计算代价:低大面积覆盖成本:目前无法实现4、VIO/SLAM(Hololens 和 Tango)这两者都是通过深度传感器的数据,通过称为Visual-inertial Odometry的方法来计算设备的相对6DOF位置和旋转。这一点要非常明确,仅仅是相对的位置。VIO的方法也很成熟了, 主要有两条路,一条是用图像信息,最近很火的Hovercam和大疆Phanton4 都是这类,而Hololens和Tango用了深度信息来计算VIO。顺便提一下暗潮涌动的深度传感器。最早成名的是Primesense被苹果收了,2015年开始停止授权,苹果憋大招不明觉厉。Pebbles Interfaces被小米投了;Intel收了softkinetic出了Realsense;微软收了Canesta出了kinect2;Google买了Movidius用在Tango上。这是个你有我必须有的传感器,而微软和Google都运用到了定位和VR/MR上。这些深度传感器基本都是Structured Light,用ToF的比较少,我后面会提一下ToF。这篇文章可以帮助理解。为什么很多人称之为SLAM,那是机器人和三维扫描里用的技术,用来建模环境。我们很久以前写过用SLAM的算法,后来做VR就从中间抽出VIO的部分就可以用来做定位了,也给自家投资人展示过。VIO就是下图中框内部分,获得6DOF就可以停止了,不需要再Register 点云了。我简单讲一下深度信息的VIO方法,深度传感器获得一般是 RGBD信息,比普通相机的信息的RGB多一个深度信息。一般光学算法直接二值化(threshold)了,RGBD则一般不可以。如果按256色,深度也是256级算,数据量是光学算法的8000多倍。为了减少计算量,会用SIFT+ RANSAC在RGB信息里找到这一帧和上一帧之间的特征点。然后再用深度信息比较特征点的空间位置,获得两帧之间的6DOF变化。因此准确的说,VIO获得的是d6DOF,是变化值。图来自eric-yuan.meSIFT(Scale-invariant feature transform)是寻找特征(feature)的方法。发布于2004年,2年后,SURF(Speeded Up Robust Features) 发布。SIFT和SURF是目前最常用的特征算法,用数学方法来理解图像的重要基础。上图中有很多正确的特征点匹配,也有很多错误的,基本上交叉的线都是错误的。RANSAC基本原理就是随机选择其中3对点,计算homography(这个词实在不知道中文,意思是对于同一画面不同相机/不同相机位置之间矩阵,一般是R|T)。如果然后收集所有对点的计算结果,计算出consensus(最多接近的homography)。肉眼看,那些平行线应该就是。前面提到,SIFT/SURF和RANSAC都运用于RGB信息,刚上手可以从OpenCV开始,这几个算法都包括,完全够用了。然后把特征点通过深度信息计算为点云,计算homography就有6DOF的变化了。如果继续想要SLAM,如前面提到的需要registered,然后写成polygon,这部分可以学习一下PCL库,跟深度、点云有关的最基础算法PCL都提供了。VIO的最大好处是不需要外部设备帮助,因此面积成本为0。但是,Johnny Lee自己也承认,VIO有两个主要技术缺陷:1、&每次开机它不知道设备的位置。2、&长距离和长时间的使用,误差会累计变成drift。微软没有提Hololens怎么解决这两个缺陷。但是Johnny Lee一直在开发Area Learning,通过用数学描述记住环境信息,然后通过辨认环境来确定自己的开机位置和矫正长时间误差。用数学描述环境这不就是我们之前特意提到的Haar feature嘛。这个方法是否有效还不能判断,但是毕竟是世界第一大牛在做,也许能出奇迹。从理论推导和实验结果我们都判断即有了J Lee的Area learning ,VIO的方法在两个情况下无法工作:1、&面对环境不可辨认,比如纯色的墙,纯色的桌子。桌子越乱定位效果越好,处女座比较惨。2、&面对室外环境。因为深度传感器有距离限制,空间大小超出限制就没有深度信息了。我们看到Hololens的demo把地板贴到墙上来做VIO,如果是白墙这个demo就无法工作了。因此我们停止了VIO用于定位的开发,但是,深度依然非常好玩,在VR应用中也许会起到意料之外的作用。总结:6DOF:可以单个成本:最高(深度,视觉传感器)计算代价:最高(CPU配合GPU计算)大面积覆盖成本:0,但是某些场景无法工作5、ToF (UWB,光,声)ToF是按照时间来计算的。基站发出信息t0,你收到信号的时间t1,用信号的飞行速度v*dt就知道了离这个基站的距离,而多个预先知道位置的基站就可以提供三角定位。这和GPS的原理接近。ToF用时间来计算,因此精度特别取决于处理器主频。我们知道光和UWB 的飞行速度是3E11 mm/s。如果我们用4.8Ghz的的主频,两个周期之间的时间,光可以飞行62.5mm,如果用20M hz的ATmega,每个周期光可以飞行15000mm。这个数字就是计算精度。如果我们需要mm级的精度,只有提高主频,或者降低信号的飞行速度。同样是20M hz主频,用音速的话,每个周期声音飞行0.017mm,这就是可以定位了。但是声音的矫正和信号编制是个挑战,所以基本上ToF的方案就不考虑了。▌哪种定位技术最好?我们前面提到我们的方法来分类的话,AR是不需要定位的,VR和MR本质是一种技术,VR是100%的MR,VR依然比MR对定位精度的要求会略低一点,而MR能用的VR一定能用。我们把刚才所讲到的几种定位方法总结成一张表:除了IMU不可以,后三种技术都可以独立为VR设备定位,而一般都会用IMU数据做辅助矫正。其中,光学定位和VIO技术目前就可以大范围使用。光学覆盖成本很高,VIO设备成本很高,而且某些场景下会失效。因此 目前没有一种能满足你的要求。以下都是我们一家之言了,我们长期关注相关的研究论文,这几年VR的高速发展不同团队也已经几乎穷举了所有基础研究的成果了,存在被忽视错过的基础方法的可能性不高。前面我特意保留了磁力方案和声音ToF方案,也许这些看似奇葩的方案会是将来的主流,但是它们是彻底死胡同的可能性非常高。从另一个角度说,技术本质是一种交易,用成本换速度,用能耗换时间。深度学习、无人车等平行市场的需求推动计算构架的高速发展,Intel、ARM,Nvidia的CUDA,以及Movidious的Myriad 2(DJi4 和Tango用了它)都有可能给我们有更多筹码去做性能上的交易。因此,可以先放弃低功耗,低性能要求这两个条件,去探索如果功耗性能都不做限制会不会有方案。而对于VIO/SLAM我们会建议保持算法更新,然后一只眼睛看微软,一只眼睛看Tango,万一在理论上有突破也可以很快出产品。这个部分我不认为是硬件问题,是一个纯软件算法的方法。我个人不太崇尚有明显缺陷的牛逼技术,比如Siri,只要有10%的话听错就无法被大范围使用,因此不是特别看好VIO。但是因为SLAM使用场景很大,机器人ROS等都可以用,写一个不会吃亏。我们团队除了自身的技术以外,最关注的是Lighthouse,我前面解释了,它几乎可以满足所有需求,但是不能大面积铺设。也许光通信的某种可能被忽略了,也许可以用声音或者UWB扫描环境,Lighthouse离大面积使用只有一个假设距离,而其他技术似乎还有好几个假设的距离。我们重视Lighthouse还有一个原因,如果它能够被大范围使用,它的铺设成本是线性的,这个和光学大面积的指数级关系会有巨大的成本差别,完全有可能成为室内的GPS。和VIO相比我觉得Lighthouse更有可能有所突破。最后,当然我认为Hoevo的方案在今天是最好的。▌Hoevo做的是什么?Hoevo的整个团队包括我,深深相信VR/MR会像GPS,移动网络一样铺满这个世界,无处不在。我们相信我们有责任推动这样的未来更快实现,后来才发现我们具备必要的技术。要实现这个未来,就需要在大范围定位技术上有突破,就像我前面分析的各种目前技术的缺陷。很难想象Optitrack或者TheVoid能在每一个城市每一个商场都有,成本太高,架设调试太困难了。Hoevo的方案500元每平米,普通建筑工人都可以建设,理论上可以铺设无线范围的空间,提供高刷新率的6DOF绝对定位。Pillsbury和Sughrue两家美国顶级专利律所正在帮助我们在全球范围PCT专利的申请,我们希望能保持技术和专利领先,这样我们才能比今天更开诚布公的讨论技术。我们也完成了一套Mixed Reality的数学模型,我们希望可以很快将SDK提供给所有人,让更多人加入VR/MR的革命里来。最后是我们的demo:视频参考资料:1、UC Davis大神Oliver的视频:https://www.YouTube.com/watch?v=_q_8d0E3tDk2、Oculus CEO Brendan Iribe:VR沉浸的5个要点:http://www.roadtovr.com/oculus-shares-5-key-ingredients-for-presence-in-virtual-reality/3、Model-Based Object Pose in 25 Lines of Code:http://www.cfar.umd.edu/~daniel/daniel_papersfordownload/Pose25Lines.pdf4、SIFT(Scale-invariant feature transform)是寻找特征(feature)的方法:Lowe, David G. (2004). &Distinctive Image Features from Scale-Invariant Keypoints&. International Journal of Computer Vision 60 (2): 91–110. doi:10.1023/B:VISI..99615.94.5、SURF(Speeded Up Robust Features)&:Bay, H., Tuytelaars, T., Van Gool, L., &SURF: Speeded Up Robust Features&, Proceedings of the ninth European Conference on Computer Vision, May 2006.6、RANSAC: Martin A. Fischler & Robert C. Bolles (June 1981). &Random Sample Consensus: A Paradigm for Model Fitting with Applications to Image Analysis and Automated Cartography& (PDF). Comm. of the ACM 24 (6): 381–395. doi:10..358692. &7、SLAM比较简单好用的论文:F. Endres, J. Hess, N. Engelhard, J. Sturm, D. Cremers, W. Burgard: An Evaluation of the RGB-D SLAM System, Proc. of the IEEE Int. Conf. on Robotics and Automation (ICRA), 2012
创业者们,如果你或你的朋友想被优派网报道,请狠戳这里&&&&
文章总数0万总阅读量 最新文章
微信扫描,关注心理学与生活,在这里找到你内心的答案!
客服邮箱:
APP推广服务:
优派QQ2群():
优派QQ1群():

我要回帖

更多关于 ProjectZero 的文章

 

随机推荐