---
格式版本: 2
标题: "Disney’s Olaf: From the Screen to Reality via Physical AI S81492 | GTC San Jose 2026"
原文链接: "https://www.nvidia.com/gtc/session-catalog/sessions/gtc26-s81492/"
发布日期: "2026-08-14"
发布时间校准状态: "found"
发布时间需复核: "否"
发布时间来源: "rule:local:strict_original_body"
发布时间证据: "Published Time: Fri, 14 Aug 2026 17:26:04 GMT"
发布时间校准原因: "规则确认唯一严格发布时间，来源 local:strict_original_body"
发布时间校准置信度: "high"
发布时间候选数量: 4
发布时间严格候选数量: 1
发布时间原页读取状态: "source template page reused from URL open"
发布时间未找到原因: ""
发布时间校准时间: "2026-08-16T01:43:41+08:00"
发布时间仲裁状态: "skipped"
发布时间仲裁尝试次数: 0
发布时间仲裁耗时毫秒: 0
发现时间: "2026-08-16T01:40:07+08:00"
入库时间: "2026-08-15T17:43:41.288Z"
来源平台: "固定入口"
搜索渠道: "fixed_url"
搜索词: "https://www.nvidia.com/gtc/"
匹配关键词:
- "GPU"
- "performance"
- "AI"
相关厂家:
- "NVIDIA"
- "Google"
相关专家:
[]
内容类型: "网页"
抓取工具: "Jina Reader"
清洗工具: "Jina Reader Markdown + Defuddle/Readability 正文提取"
原始附件:
[]
AI优质: "否"
AI打分: 10
AI分档: "非优质"
AI质检状态: "不通过"
AI打分理由: "正文为迪士尼Olaf机器人物理AI演讲，与超节点/AI Rack/机柜级AI基础设施无关，主题完全不相关。"
AI质检模型: "ali-deepseek-v4-flash"
AI质检时间: "2026-08-16T01:43:50+08:00"
AI主题相关性: 0
AI来源权威性: 10
AI新颖性: 0
AI技术细节: 0
AI商业部署信号: 0
AI完整性: 0
AI摘要: "迪士尼Imagineering团队在NVIDIA GTC 2026上介绍机器人角色Olaf及BDX机器人，利用强化学习和模块化硬件，并搭载NVIDIA Jetson，将屏幕角色带入乐园现实。"
AI摘要模型: "ali-deepseek-v4-flash"
AI摘要时间: "2026-09-07T03:30:31.286Z"
采集批次: "2026年8月15日21点37分12秒"
采集批次ID: "20260815-213712-121"
去重键: "https://www.nvidia.com/gtc/session-catalog/sessions/gtc26-s81492"
---
Title: Disney’s Olaf: From the Screen to Reality via Physical AI S81492 | GTC San Jose 2026
URL Source: https://www.nvidia.com/gtc/session-catalog/sessions/gtc26-s81492/
Published Time: Fri, 14 Aug 2026 17:26:04 GMT
Markdown Content:
Visit your regional NVIDIA website for local content, pricing, and where to buy partners specific to your country.
[Continue](https://www.nvidia.com/)
[**GTC Berlin** October 20–22](https://www.nvidia.com/en-eu/gtc/) | [**GTC 2027** March 15–18](https://www.nvidia.com/gtc/)
* [](https://www.nvidia.com/gtc/session-catalog/sessions/gtc26-s81492/#)
* [](https://www.nvidia.com/gtc/session-catalog/sessions/gtc26-s81492/#)
* [](https://www.nvidia.com/en-us/account/)
* [Log In](https://www.nvidia.com/gtc/session-catalog/sessions/gtc26-s81492/#)[LogOut](https://www.nvidia.com/gtc/session-catalog/sessions/gtc26-s81492/#)
* [EN](https://www.nvidia.com/gtc/session-catalog/sessions/gtc26-s81492/#)
* [EN](https://www.nvidia.com/en-us/on-demand/)
* [简中](https://www.nvidia.cn/on-demand/)
* [日本語](https://www.nvidia.com/ja-jp/on-demand/)
* [한국어](https://www.nvidia.com/ko-kr/on-demand/)
* [繁中](https://www.nvidia.com/zh-tw/on-demand/)
PLATFORMS
other links
[](https://www.nvidia.com/gtc/)
Keynote
* [Keynote](https://www.nvidia.com/gtc/keynote/)
* [_GTC Live_ Pregame](https://www.nvidia.com/gtc/pregame/)
Explore
* [Conference Topics](https://www.nvidia.com/gtc/conference-topics/)
* [Poster Gallery](https://www.nvidia.com/gtc/posters/)
* [Speakers](https://www.nvidia.com/gtc/speakers/)
* [Startups & VCs](https://www.nvidia.com/gtc/startups/)
* [Workshops, Training Labs & Certification](https://www.nvidia.com/gtc/training/)
[Sponsors & Exhibitors](https://www.nvidia.com/gtc/sponsors/)
More
* [Code of Conduct](https://www.nvidia.com/gtc/code-of-conduct/)
* [Contact Us](https://www.nvidia.com/gtc/contact/)
* [FAQ](https://www.nvidia.com/gtc/faq/)
* [Privacy Policy](https://www.nvidia.com/en-us/about-nvidia/privacy-policy/)
* [See All GTC Events](https://www.nvidia.com/gtc/events/)
* Keynote
* [Keynote](https://www.nvidia.com/gtc/keynote/)
* [_GTC Live_ Pregame](https://www.nvidia.com/gtc/pregame/)
* Explore
* [Conference Topics](https://www.nvidia.com/gtc/conference-topics/)
* [Poster Gallery](https://www.nvidia.com/gtc/posters/)
* [Speakers](https://www.nvidia.com/gtc/speakers/)
* [Startups & VCs](https://www.nvidia.com/gtc/startups/)
* [Workshops, Training Labs & Certification](https://www.nvidia.com/gtc/training/)
* [Sponsors & Exhibitors](https://www.nvidia.com/gtc/sponsors/)
* More
* [Code of Conduct](https://www.nvidia.com/gtc/code-of-conduct/)
* [Contact Us](https://www.nvidia.com/gtc/contact/)
* [FAQ](https://www.nvidia.com/gtc/faq/)
* [Privacy Policy](https://www.nvidia.com/en-us/about-nvidia/privacy-policy/)
* [See All GTC Events](https://www.nvidia.com/gtc/events/)
* [](https://www.nvidia.com/gtc/session-catalog/sessions/gtc26-s81492/#)
* [EN](https://www.nvidia.com/gtc/session-catalog/sessions/gtc26-s81492/#)
* [EN](https://www.nvidia.com/en-us/on-demand/)
* [简中](https://www.nvidia.cn/on-demand/)
* [日本語](https://www.nvidia.com/ja-jp/on-demand/)
* [한국어](https://www.nvidia.com/ko-kr/on-demand/)
* [繁中](https://www.nvidia.com/zh-tw/on-demand/)
[Watch On Demand](https://www.nvidia.com/en-us/on-demand/search/?facet.event_name[]=GTC%20San%20Jose&facet.event_year[]=2026&facet.mimetype[]=event%20session&headerText=All%20Sessions&layout=list&page=1&q=-&sort=relevance&sortDir=desc>cnavinherit=true)[](https://www.nvidia.com/gtc/session-catalog/sessions/gtc26-s81492/# "Menu")
[Watch On Demand](https://www.nvidia.com/en-us/on-demand/search/?facet.event_name[]=GTC%20San%20Jose&facet.event_year[]=2026&facet.mimetype[]=event%20session&headerText=All%20Sessions&layout=list&page=1&q=-&sort=relevance&sortDir=desc>cnavinherit=true)[](https://www.nvidia.com/gtc/session-catalog/sessions/gtc26-s81492/# "Menu")
* [Keynote](https://www.nvidia.com/gtc/session-catalog/sessions/gtc26-s81492/#)
* [Keynote](https://www.nvidia.com/gtc/session-catalog/sessions/gtc26-s81492/#)
* [Keynote](https://www.nvidia.com/gtc/keynote/)
* [_GTC Live_ Pregame](https://www.nvidia.com/gtc/pregame/)
* [Explore](https://www.nvidia.com/gtc/session-catalog/sessions/gtc26-s81492/#)
* [Explore](https://www.nvidia.com/gtc/session-catalog/sessions/gtc26-s81492/#)
* [Conference Topics](https://www.nvidia.com/gtc/conference-topics/)
* [Poster Gallery](https://www.nvidia.com/gtc/posters/)
* [Speakers](https://www.nvidia.com/gtc/speakers/)
* [Startups & VCs](https://www.nvidia.com/gtc/startups/)
* [Workshops, Training Labs & Certification](https://www.nvidia.com/gtc/training/)
* [Sponsors & Exhibitors](https://www.nvidia.com/gtc/sponsors/)
* [More](https://www.nvidia.com/gtc/session-catalog/sessions/gtc26-s81492/#)
* [More](https://www.nvidia.com/gtc/session-catalog/sessions/gtc26-s81492/#)
* [Code of Conduct](https://www.nvidia.com/gtc/code-of-conduct/)
* [Contact Us](https://www.nvidia.com/gtc/contact/)
* [FAQ](https://www.nvidia.com/gtc/faq/)
* [Privacy Policy](https://www.nvidia.com/en-us/about-nvidia/privacy-policy/)
* [See All GTC Events](https://www.nvidia.com/gtc/events/)
[Video 1](blob:https://www.nvidia.com/81e39b1b-e638-4026-984f-9ca167590803)
Play
Play
Play on TV
00:00
Play
Seek 10 seconds backwards
Seek 10 seconds forward
00:00 / 40:32
Mute
Press TAB to open volume control Use the arrows to control the volume
Enable Captions
Settings
Play on TV
Turn on Picture in picture
Show Full screen
Transcript Powered by AI
X
X
00:12
Good morning.
00:15
Yeah, thanks so much, Akhil.
00:17
It's the third year in a row that we get to present here,
00:21
and it's very humbling to be able to do that.
00:25
And we have a really amazing partnership with NVIDIA, so
00:27
I'll highlight some of the work that we actually do together.
00:31
So I'm going to talk about OLAF, but I'm also going to talk about
00:35
robotic characters in general.
00:37
So it's important that our technology that we build is
00:40
basically... Oh, could we go back?
00:42
Alright, it's advancing.
00:46
One more back.
00:50
So you're going to talk about technology behind robotic
00:54
characters, technology that helps, that is agnostic to
00:57
the character that we build and helps to build them quickly.
01:01
So the theme here is to go from screen to reality, and
01:05
we use physical AI, and physical AI means actually something different
01:08
for the Walt Disney Company, so I'll explain what that means.
01:13
All right, so a robotic character, what is a robotic character?
01:16
So some of you might have seen a robotic character before,
01:20
definitely yesterday some of you have seen, or most
01:24
of you have seen OLAF.
01:26
But so let me briefly explain what a robotic character is and
01:28
let me use our BDX druids to do so.
01:32
So BDX druids are new to us because they can freely roam in our parks.
01:38
They are bipedal systems, they have actually five degrees
01:41
of freedom in each leg.
01:43
So they have one degree of freedom less than we humans do.
01:47
So walking is more challenging for them.
01:50
But what's unique about them is that they can walk
01:52
around, interact with objects, interact with our guests.
01:57
And you as a guest can co-direct a story, so you can influence a
02:01
story, and every time you meet such a druid, the story is different.
02:06
So they enable us at The Walt Disney Company to augment our lens.
02:13
All right, so some of you might know that already, but
02:16
the inspiration for the BDX droids was the BD-1.
02:20
So this character took actually first steps in a game, not
02:26
in a movie, so in a game.
02:28
It's an explorer droid that assists basically if you're
02:31
on a mission to explore the area, and it's also a companion.
02:36
And so that served as an inspiration for us
02:38
to do the BDX Druid.
02:39
You see clear differences here, but so the game character
02:43
is basically smaller.
02:44
We had some constraints, actuator sizes and so on, and thought
02:48
we could just build an adult version of a BD type Druid.
02:53
If you think of a puppy, you know, the BDX would be the adult version.
02:59
But so since we last talked, since we were here last year,
03:03
we brought actually the VDX droids for limited engagements
03:06
to all of our global parks.
03:09
So this is really exciting for my team and all of us
03:12
at Imagineering to bring these droids to all of our parks.
03:16
And even we brought them to a cruise ship, the first
03:19
cruise ship, the Vish.
03:22
The good thing is they're going to come back every now and
03:25
then to all of these global parks.
03:28
This is thrilling to us.
03:30
All right. But now a bit more about the technology.
03:33
So we built this technology or tools that help us to build
03:37
these robotic characters, our new robotic characters quickly.
03:40
So we want to be able to build custom characters quickly.
03:43
So that's a unique challenge.
03:45
And we want to do this within months, not years.
03:49
So let me explain how that works.
03:51
So at the core of this is this triangle.
03:54
So you have mechatronics design, so we use CAD software to
03:57
design our robots.
04:00
And then we can extract simulation representations from them.
04:03
And then what's unique about our use case is that we have
04:06
always artistic interfaces.
04:09
So we use animation, basically motion targets that we want
04:14
these robots to be able to do.
04:17
And then, what connects these two disciplines is
04:20
reinforcement learning.
04:21
So reinforcement learning is really bringing this together
04:24
and allows us to iterate quickly.
04:26
So let me briefly explain that and use the BDX druids for that.
04:30
So to build custom robots quickly, we use modular hardware components.
04:34
So we have actuators or a family of actuators, and we also have
04:38
sensors and a family of sensors.
04:41
And we understand these actuators and sensors well in our software
04:45
stack, meaning in the runtime stack, but also in the offline
04:48
tools that we use, for example, for reinforcement learning.
04:52
That's crucial.
04:53
You need to understand your hardware well.
04:56
And then you also have the brains and power on board, so we have a
04:58
Jetson on board, that's incredibly important for autonomy, so that's
05:02
a connection to NVIDIA, we actually get help from NVIDIA also, when
05:07
it comes to runtime systems and so on, and we have a battery on board
05:11
that lasts for about two hours.
05:14
Before you have to recharge these components are agnostic
05:17
to the character but so one thing that is specific to the
05:20
character are the so-called show functions so we don't want to just
05:24
build functional robots we want to build robots that perform on
05:28
stage so we have speakers on board a headlamp for scanning we have
05:32
eyes that we can animate and we have antennas that we can actuate
05:36
And so speakers are quite important because you want to hear
05:39
a druid and not a mechanical robot.
05:42
But those are specific to the character, rest is agnostic
05:45
to the character.
05:47
So animation is an important key ingredient.
05:50
We create technology that always has a creative interface.
05:54
We want to make sure that our amazing world talent when it comes
05:59
to animation can create content for these robotic characters.
06:04
So animation is key.
06:05
And so the goal for our robots is basically to be able to
06:09
control them, remote control them, similar to how you control a game
06:13
character, a virtual character.
06:15
So we can use procedural animation to create the target motions
06:20
that we want these robots to be able to do.
06:23
And so you want to be able to control, for example,
06:25
the VDX Droids with two joystick controllers.
06:28
So with one you control the direction of walking and with
06:31
the other one the relative motion of the head relative to the body.
06:36
But this, if you would run this in a robot, the robot
06:38
would actually fall over, because this is kinematic input, it's not
06:42
physics informed, and that's where reinforcement learning comes in.
06:47
So with reinforcement learning you can connect these two worlds, you
06:49
get the simulation representation from the mechatronics design
06:53
and the animation input and then you can basically train
06:57
so-called control policies.
06:59
And that's actually quite a simple process, so you bring
07:02
your druid or your robot into a simulation environment.
07:07
And then you have two inputs here.
07:09
You want to be able to control the Droids, that's where the
07:12
operator commands come in.
07:14
They are inputs to a simple neural network,
07:17
a so-called control policy.
07:20
You also have the robot state as an input, and the policy
07:23
is a simple functional mapping that then goes from inputs
07:26
to outputs, and outputs in this context are actuator commands.
07:32
And so, we use reinforcement learning to train those, but
07:35
what's unique about our use case is that we send the same operator
07:39
commands during training also to an animation engine to define a
07:44
target state for every robot state, and then we can define rewards.
07:49
If you can follow an animation well, then you get high rewards,
07:53
especially also if the robot doesn't fall over.
07:55
But if the robot falls over, or if you don't follow these
07:58
stylized motions well, you get a lower reward.
08:01
And then reinforcement learning simply updates this mapping
08:04
by updating the weights.
08:08
And the important thing is that we randomize these commands.
08:11
So if you don't do that, there could be configurations or
08:16
combinations of the two joysticks or buttons and so on where
08:19
the robot could fail.
08:20
So you want to sample that entire space during training
08:23
so that you as an operator can do arbitrary combinations
08:26
of the operator commands.
08:28
While the robot is always stably balancing.
08:32
So here is the process, so we initialize these weights
08:35
basically with random numbers.
08:37
At the beginning the robot falls over, but over time
08:40
if you continue to update these weights, it not only balances
08:44
better, but slowly but surely can do the stylized reference motions
08:49
that our artists are interested in.
08:53
And so it's incredible how well this works and scales also.
08:56
That's again where GPUs come in.
09:00
So we use NVIDIA hardware for training here.
09:03
You can train so-called policies, these policies, you know,
09:07
in hours to a couple of days.
09:10
And we vary here the terrain.
09:12
We randomize the terrain so that they can stably balance even
09:15
if you don't have flagged ground.
09:17
And we also push them.
09:18
So we apply forces and torques that you see here in green and blue.
09:22
So that they know how to recover, do recovery steps
09:25
and don't fall over.
09:26
That's where the robustness comes from.
09:29
And so what amounts to a decade of training you can do in a
09:32
few hours, today's on-desktop GPUs.
09:35
Amazing, I think, what that unlocks.
09:38
Okay, so that gives you a really nice abstraction layer, so here...
09:42
You know, independent of what an operator does, even if you have
09:46
a very challenging environment, this robot learned to balance
09:50
through this process by randomizing the simulation, basically.
09:54
All right, so that opens up the question, well, you have
09:57
now an abstraction layer.
09:59
How can you make these characters more autonomous?
10:01
And I want to give you an update on our autonomous or
10:04
autonomous solutions there.
10:06
And so for us, it's not just about functional autonomy,
10:09
it's about believable autonomy.
10:11
You want to believe that this is a character on stage, not just
10:15
functionally going from A to B.
10:18
So one item that you can think about in an autonomy stack is
10:23
navigation and so I just explained that this you know abstraction
10:27
layer works quite well where you control remotely a droid with
10:32
operator commands so it abstracts basically the lower level controls
10:36
from a user but now the question is could you actually replace
10:40
this with a navigation policy?
10:44
And that takes from a brain navigation commands, for
10:47
example, waypoints along a trajectory as an input.
10:50
And then you want to be able to follow this trajectory,
10:53
while obviously you don't want to bump into anything
10:56
and you want to be in character.
10:58
And so to be able to do that, you not only need the robot
11:00
state, you also need the state of the environment.
11:03
You need perception models.
11:06
So you need a sensor model.
11:08
So a sensor model then takes the simulated environment
11:13
as an input and provides a sensor state that then serves as
11:17
an input in its navigation policy.
11:21
So then we apply actually something very similar to what
11:24
we do if we train a control policy.
11:27
We send the same navigation commands to an artistic goal
11:30
engine that provides us then with a target state.
11:34
And this target state could be discrete, that discrete
11:36
time steps you want to be at a certain location.
11:40
That's sort of a simple case, and could be a partial state
11:43
of the robot, for example, just the pelvis, or it could
11:46
be the full state, and it could be dense inputs, because we want
11:49
to have a robot not just go from A to B, but do this in style, with
11:54
a personality, and maybe that's not the straight path, actually.
11:58
But so you can then compare the target state to the robot state
12:01
again, define rewards, and use reinforcement learning to update
12:06
those weights, While you freeze, basically, the control policies.
12:11
Again, it's important to always randomize your commands,
12:13
otherwise your system is not going to be robust.
12:16
That's where robustness comes from.
12:18
So, for example, you randomly choose waypoints that
12:22
you want to reach.
12:24
So let me show an example of this in action.
12:26
So here you see druids that learn to navigate.
12:30
You see the sensor simulated, and they give these target
12:33
points, the green points, and they learn basically to
12:36
get to these target points without bumping into any obstacles.
12:40
And so we do that by also procedurally changing the
12:43
obstacles or make it really challenging environments.
12:48
And so we also brought this to our robots of use.
12:51
So everything that you see here runs on a Jetson on
12:54
board of the robot.
12:55
We stream this video data that you see in the inset.
12:59
And the robot, online, does mapping, so it can map the
13:03
environment, it sees obstacles.
13:05
So if you didn't explore an area yet, it's an explorer droid, right?
13:09
Then you see white, but as soon as you see obstacles, it updates
13:13
the map and, you know, every voxel also has a confidence value, how
13:17
confident the robot is that there is an obstacle or no obstacle.
13:22
And so you can challenge the robots with these obstacles,
13:26
so that's Dario and Sami working on this part of the project.
13:32
And so you can make it really challenging and you see that
13:34
the robot learns to not bump into anything, it always goes
13:38
around the obstacles, and it does that in a quite natural way.
13:42
But the robot also does usually this scanning every now and
13:45
then, that's part of the story, it's an explorer droid, so things
13:48
like that are important to us.
13:51
All right, so here is another autonomy task
13:55
for us, human-robot interaction or guest-robot interaction.
14:00
And so here we had an amazing operator who really knows how to
14:03
get the best out of the character.
14:06
So how can you get to human-operator level autonomy with
14:11
a completely autonomous solution?
14:14
That's a huge challenge.
14:15
And I think that's a challenge in robotics in general, not
14:17
just for robotic characters Those are our use case.
14:21
So what you can do is we can record sensor data together
14:26
before an operator does.
14:30
That's a hugely important and valuable data set
14:35
that it can capture.
14:37
And so if you have these data sets captured, you
14:40
can basically train a network.
14:42
We use a diffusion model at this moment in time to imitate
14:46
what the operator does.
14:47
So you can then go and map sensor states to operator commands.
14:53
And so that's how we can actually achieve quite believable autonomy.
14:57
That's shown here for a robot that is shy, so we ask operators
15:02
to author content for a shy robot.
15:06
So it doesn't, it does never look towards Sammy, it tries
15:09
to look away, it says go away.
15:12
So it's remarkable how well this works, and you know, you need only
15:17
very little data for that, so not a huge, huge data set, so that was
15:21
quite interesting for us to see.
15:24
So really the personality comes out.
15:28
Alright, so the BDX druids were special as characters
15:32
because they took first steps in a park and now are going to appear
15:37
in the Mandalorian Grogu movie.
15:39
So usually if you go the other way around, we start in a
15:41
film or in a movie and bring these characters to life in the park.
15:45
And so that's actually a simpler problem to solve because robotic
15:48
characters, so we built literally a robotic character, a druid,
15:53
that's easier to build than trying to get other characters
15:57
to life in the physical world.
16:01
So, then we started working on OLAF.
16:03
And OLAF is an animated character.
16:05
Animated characters can do very non-physical things.
16:08
They don't have to observe the laws of physics.
16:12
For example, OLAF has a gigantic head, right, and has only two
16:15
snowballs sticking out for walking.
16:17
So how can you achieve something like that in the physical world?
16:21
That was a huge challenge to us.
16:23
At the beginning, I was actually pretty worried that we won't
16:26
succeed in building OLAF so that it's a high-quality,
16:30
Disney-quality experience.
16:33
But I think we got there.
16:35
And we got there within months.
16:36
So here you see OLAF taking steps at the World of Frozen in Paris.
16:42
This is gonna open up next week.
16:44
So you'll be able to see OLAF at the park
16:47
at Disneyland Paris next week.
16:50
So that was an amazing experience and it's a fantastic world
16:53
also for OLAF to walk around.
16:58
And so we brought, or are going to bring, OLAF to Disneyland Paris
17:02
and then soon after to also our Hong Kong park where there is an
17:06
existing World of Frozen already.
17:09
And so now I'm going to introduce you to a very
17:12
special guest, OLAF himself.
17:16
Hi, I'm OLAF and I like warm welcomes.
17:26
Good morning.
17:32
Look at us, out on the town.
17:34
I'm wearing my fancy kohl.
17:37
Yeah, also I iron my shirt, Olaf.
17:40
So you're quite a celebrity by now, huh?
17:42
Alright, nothing to see here.
17:44
Just an enchanted snowman walking among you.
17:48
So what has your favorite moment been at GTC so far?
17:53
Good question.
17:55
Um, maybe, nope, I can't narrow it down.
18:00
It's like asking me to choose my favorite snowflake.
18:04
So do you have any advice for us before we continue
18:07
with the presentation?
18:08
Do every possible thing to make friends with every single
18:11
person you meet the entire day.
18:13
Are you taking notes?
18:15
Yeah, noted. I'm gonna follow your advice, OLAF.
18:19
So thanks so much. We'll continue with the presentation.
18:21
You can come back later.
18:23
Goodbye, friends.
18:25
I'll never forget most of you.
18:34
Now, now, no need for applause, but no need for not applause either.
18:43
Thank you. Thank you.
18:49
Alright, now it's difficult to continue because you will
18:53
probably want to see OLAF more, but I'm gonna try.
18:56
Bye for now!
18:59
So here you see the mechatronics design, we can see inside
19:04
of OLAF and we actually use two legs that are exactly the same and
19:08
mounted them in opposite, you know, mounted them so that one knee bends
19:13
forward, the other one backward.
19:15
So for the arms we used so-called spherical linkages and also
19:19
in the head we used linkages.
19:21
And, you know, traditionally if you look at humanoid
19:24
characters you usually have actuators at the joints.
19:28
And so those systems are really hard to control.
19:32
So there is innovation in the leg design to maximize
19:35
the workspace, but then also we used many linkages to place
19:38
actuators where there is space.
19:41
For example, for the arms, we placed the actuators, two actuators
19:45
inside, and then the spherical linkage allowed us to really,
19:49
you know, move these like sticks that stick out of the snowman
19:53
without you seeing the actuators.
19:55
So that was a huge challenge for us.
19:58
So we use actually the same modular hardware components
20:00
like we used for BDX largely.
20:03
We added a couple of additional actuators to the actuator family,
20:06
characterized them, so we have good simulation representations, but
20:10
we used largely the same hardware.
20:12
And again, a Jetson is on board, we also have a battery on board that
20:16
lasts actually a bit less long than for the BDX druids, but long enough
20:22
for doing several shows in a row.
20:24
And show functions, if you have speakers on board, you
20:26
can also plug in to audio systems, as you heard, directly stream
20:32
the audio from the robot itself.
20:35
So model hardware is important, and for reinforcement learning,
20:37
we actually use this framework that I explained at the very
20:41
beginning for the BDX-Druids.
20:44
But there was a challenge, and the challenge was simulation,
20:48
or e-simulation in this context.
20:51
Why is this challenging?
20:52
Well, we have good simulators out there for articulated body systems.
20:56
Most simulators that are fast, GPU-accelerated, are made
20:59
for articulated body systems.
21:02
But so OLAF has these spherical linkages, has
21:05
also linkages in his head.
21:08
So you see the articulated body part here in green and the closed
21:11
loops or these linkages in blue.
21:15
And so the linkages are challenging to control with
21:17
control policy or simulate.
21:21
So we started working on a simulator.
21:24
So most simulators that you can use for reinforcement
21:28
learning, again, they interface well with articulated body systems.
21:33
Sometimes you have loop-close constraints that it can add,
21:36
but the simulators are made for more articulated body systems.
21:39
That's just not enough.
21:40
But the simulators are good on the GPUs.
21:42
For example, Isaac Sim is a good simulator that we've
21:47
used for the BDX druids.
21:49
And they support contact also.
21:51
Contact is important for walking characters or characters or
21:55
robots that manipulate the objects.
21:58
But then, so we have built expertise at DCM Research to
22:03
build simulators that can simulate so-called audio animatronics well.
22:08
And so systems are highly complex systems that have
22:10
many closed loops and use, you know, all sorts of joint types.
22:15
And so this is again to place actuators where there's space
22:19
and remote actuators.
22:21
And so for this particular animatronic fear that is in
22:25
E1000, we have 45 so-called loops.
22:29
And these are highly complex systems that existing simulators
22:33
struggle to simulate.
22:36
So we had these simulators, but they were targeting the
22:39
CPU, single instances, and we didn't have contact because
22:43
usually these systems act on stage.
22:46
And so then we started, as part of the Newton framework,
22:50
the collaboration with NVIDIA and Google DeepMind, we started working
22:54
on a Camino simulator that brings these two worlds together, supports
22:58
these more general mechanisms with these closed-loop systems, and does
23:01
this on a fundamental level, is GPU-accelerated and allows contact.
23:09
So the name Camino comes from the Star Wars planet, where the
23:14
inhabitants are experts in cloning.
23:15
So you clone many robots in simulation, that's why we
23:18
thought that name makes sense.
23:22
And again, the Camino simulator, the ultimate goal of that
23:25
simulator is to be able to apply reinforcement learning
23:28
to control any mechanism.
23:31
So that's the goal.
23:33
Very complex mechanisms, too.
23:35
And OLAF was just the first one we are going to use
23:38
this simulator for.
23:41
So if you want to learn more about Newton, our collaboration
23:43
with NVIDIA and DeepMind and Dear Progress, together with
23:47
our progress on Camino, come to this session on Newton on Thursday.
23:53
So we'll do a more deep dive there.
23:56
So ultimately, once you have a simulator, you can apply
23:59
reinforcement learning and follow these artistic motions well.
24:06
And so you can really get close to these artistic intents.
24:08
So you see the animation reference in the inset video here and
24:12
how close you can get with the two snowballs for walking.
24:17
So here is another reference animation.
24:20
And so again, the match between what's physically happening
24:25
to what the artist wanted, it's a perfect match.
24:28
You can't see the differences any longer.
24:33
Right, so one important aspect was also heel-toe walking.
24:36
So usually when you see robots walking, they always have
24:39
full contact with the ground, with the full surface of the foot.
24:44
So that didn't work for us.
24:46
Rolling off was actually quite important to do this or hit
24:49
the creative vision for this.
24:50
So we extended our tools on the procedural animation side
24:55
to achieve this goal.
24:58
And so we didn't stop innovating there.
25:00
So reinforcement learning provided us with control policies.
25:05
We were able to do first walking cycles on the physical system.
25:10
But we realized that this OLAF makes a lot of noise, too
25:15
much noise, when he walks around.
25:17
So that doesn't work, right?
25:19
The snowman doesn't do noisy walking, so we had to find
25:24
a solution for that.
25:25
And so I'm going to show you a video and I'll talk
25:27
about that video once the audio is fully played.
25:38
Too loud for snowmen.
25:43
So what we did here is we introduced so-called impact
25:46
reduction rewards so they really you know reduced the
25:51
noise that they can hear and that made a huge difference so
25:54
the second motion or the second you know video that I showed is much
26:00
much better much more believable and so if I switched off the audio
26:04
you would not see a difference it's a much more believable
26:08
motion and impact reduction is useful in general for robotics
26:12
Because if you can reduce the impact forces that act
26:16
on a mechanical system, you have less wear on the system,
26:21
especially also on your actuators, and you have to replace them less.
26:24
So it actually helps to reduce maintenance cost, or has the
26:28
potential to do so.
26:29
So what we innovated for OLAF has applications elsewhere as well.
26:34
And then the second innovation was that we introduced rewards
26:37
that help with, basically, OLAF learning to not melt.
26:43
So he loves summer, so we need to prevent him from melting.
26:48
And so we saw overheating for the actuators.
26:53
So we introduced thermal dynamics modeling.
26:55
So we not only simulate the mechanical system, but also the
26:58
thermal dynamics of the actuators to prevent overheating, make the
27:03
least amount of changes possible to stay as close as possible to the
27:08
artistic intent while preventing the system from failing due
27:11
to overheating in the actuators.
27:14
So we have additional rewards that help with that.
27:18
And so what I'm going to show you is two videos side by side.
27:22
And so OLAF has a huge head.
27:25
And the problem here is we had to fit three small actuators.
27:28
So you have all the degrees of freedom necessary to move
27:31
the head in the thin neck.
27:33
We didn't want to make any changes because it's OLAF, you can't.
27:37
You have to keep the outside the same.
27:39
So large head, even if you reduce the weights as much
27:42
as possible, we saw overheating of those actuators in the neck.
27:47
And so with those additional rewards and the thermal dynamics
27:51
modeling, we could prevent basically the actuators to
27:55
ever cross a certain temperature limit with a safety margin.
28:01
And so I'm going to show these two videos.
28:04
And so what you are going to see with thermal modeling
28:07
is that the higher the temperature,
28:13
The more we lean forward, and if the temperature is
28:17
decreasing again, the robot goes back and leans backward.
28:22
So, very subtle changes.
28:24
An operator would not perceive that, especially a guest would
28:27
not perceive that.
28:28
You would not perceive these differences.
28:31
But OLAF literally learned to manage its own heat.
28:36
And I think that's huge also in general for robotics.
28:39
Think of like robots working outdoors in outdoor environments
28:44
where it's really hot outside, so thermal modeling will
28:47
become more important.
28:49
Okay, so I want to briefly give you an update on what
28:52
we do on the research side.
28:53
We've built this humanoid character for research purposes.
28:57
And recently, we looked actually into soft and stylized falling.
29:01
And so falling is always perceived as a failure.
29:06
I don't think falling is a failure.
29:08
I think you should not exclude that from your control stack.
29:11
You should include it, because we can detect the fall and
29:15
then fall in a natural way, and in a way that doesn't
29:19
have an impact on your components.
29:21
So what you're trying to do here with a policy is
29:25
we trained a policy where you can provide a target
29:28
pose the robot should fall into.
29:31
And we had rewards to reduce the impact forces that act
29:36
onto the components so you can prevent components from failing.
29:41
So stylized falling helps also if you have a target
29:44
post the robot can fall into, you can also do more easy recovery
29:48
afterwards and in-style recovery.
29:51
So I think it's something important for robots in general,
29:54
we need to embrace that falling is just part of functionality
29:58
that we need to cover.
29:59
And it's not a failure.
30:01
So we consider that part of the character.
30:05
So here it's quite remarkable how, in a very short amount
30:09
of time, if you train a policy well enough, the robot can react to that
30:13
and fall into a particular pose.
30:14
It can even turn itself.
30:25
Okay, you can tell that we are having fun every day.
30:29
Even if the robots don't look beautiful, it's still quite
30:31
funny to work with them.
30:33
Okay, so now autonomy, how does the future look like?
30:37
Well, we have these MULTIMODAL reasoning models or MULTIMODAL
30:40
large language models that can do long horizon planning
30:44
for us and can provide us with high level commands as
30:47
an input for the next steps.
30:49
But the question is, how can you interface with those high-level
30:53
commands and how can you bring this to a robotic character
30:56
or a robot in general?
30:58
And so here we believe that this is a two-step process.
31:02
So motion diffusion models are quite useful.
31:05
We use self-supervised learning to extract from large data
31:09
sets, motion data sets, the structure to understand the
31:13
structure, understand motion.
31:16
And so video data is actually quite helpful for scalability.
31:19
In videos, you can actually observe motion, extract motion.
31:24
But so there is one bottleneck with that, and I'll talk about
31:26
that right after showing what a diffusion model does.
31:30
So a diffusion model can take high-level commands
31:32
as a conditional input, takes noise also, and denoises.
31:36
And usually we have seen that is for images, but here is
31:40
a visualization of that process, the denoising process on a robot.
31:45
And in the end, you see a very, very smooth motion.
31:48
So you could have a conditional input saying
31:50
I want this robot to do
31:53
You know, Michael Jackson moonwalk, and then with that
31:57
condition, it could perform
32:00
that motion based on learning from large data sets of motions.
32:06
But the question is, this is not physics-informed, this
32:08
is kinematic motion.
32:10
And so kinematic motion needs to somehow be mapped to the
32:14
robot itself, and for that we need reinforcement learning,
32:18
we need simulation, because what it can't observe in video
32:22
data is forces and torques.
32:24
So you need to close the gap, and that's why simulation
32:27
and reinforcement learning is so incredibly important as a tool, and
32:31
will stay relevant in this context.
32:34
So we can train them with reinforcement learning based
32:36
on this large data set of motion.
32:38
We can train a tracking policy so we can basically close the loop
32:42
fully and bring this to characters.
32:45
We've demonstrated this already a while ago, these results.
32:49
We have some more exciting things in the pipeline there.
32:53
But so I think this could basically be a pipeline that can
32:56
scale quite well and do this soon.
33:01
Alright, so this is the team in Zurich, that's the team
33:03
I get to work with, I get to present here on stage, but it's
33:07
actually those team members that should be in the spotlight here.
33:11
Everyone has a unique expertise, and without your expertise,
33:15
robots like a BDX or OLAF would not be possible.
33:19
And you also have half of the team here, so I'm going
33:22
to call them out briefly.
33:24
Agon, Dario, David, Espen and Ruben, if you could get
33:28
up, stand up briefly.
33:40
So if you have questions that you rather should ask them,
33:42
they are the true experts.
33:45
And so we also get tremendous amount of help to bring these
33:49
characters to our parks, deploy them, productize them.
33:53
There's hardware productization, there's software productization.
33:57
So it is truly a family behind this all.
34:01
Many, many, many contributors to make this possible.
34:03
So I'm very grateful that we have this support system.
34:08
These are references if you are interested in reading
34:10
up on some of this technology.
34:18
And then come to our session on Newton simulation, because
34:23
with good simulators that balance accuracy with parallelism
34:28
and speed, you can not only achieve walking with these robotic
34:33
characters, but performances.
34:35
So here is a robot doing a pirouette.
34:40
And so it's incredibly hard because frictional contact
34:42
is hard to model properly.
34:44
And so you see here stick-slip contact.
34:46
So if you wanted to learn more about simulation
34:49
and what it can help with, especially in combination
34:52
with reinforcement learning, come to our session on Thursday.
34:56
And last but not least, we have a landing page for Camino.
34:59
Camino is available now.
35:00
You can try Camino.
35:02
So you can find this landing page online, and this gives
35:05
you pointers on how to use Camino.
35:08
It's in a beta state, but we wanted to release it to get
35:12
first feedback and for you to use.
35:15
So thanks so much.
35:27
So now OLAF is here to answer your questions.
35:34
We can take a few questions..
35:43
and interactions.
35:44
You mentioned you don't need many data to train, let's
35:47
say, the shy robot.
35:48
So what's the process there to capture the emotions?
35:52
Yeah, data capture for interactions, human-robot
35:55
interaction in particular.
35:57
So the data capture, we simplified the sensors for now.
36:01
We used the mocap system to just get a sense of does that
36:05
scale at all, because we didn't know how much data you would need.
36:08
So we abstracted away the sensors with mocap.
36:11
So you saw that Sami, who worked on this project, was wearing a
36:16
hat with mocap markers on the head.
36:19
And so we didn't have full body mocap, so just on the head,
36:24
but that was actually sufficient to achieve quite good results already.
36:30
So we captured everything that an operator does, together
36:33
with mocap markers on the robot and on the human's head.
36:44
Once more, but louder.
36:49
Is there another question you could help answer?
36:53
Hello, good morning. My name is Jonathan Ramos.
36:54
I'm a GenAI team lead.
36:56
I have a question concerning the use of large language
36:59
models, particularly how did it simplify your workflow
37:03
on getting OLAF from A to B?
37:07
So, that's an excellent question.
37:08
Do you want to take this one, Olaf?
37:11
No, I'm not sure.
37:15
So at the moment OLAF is actually completely remote controlled and we
37:19
trigger recorded lines so you can actually hear Josh Gadd's voice.
37:25
So for OLAF that's going to stay that way.
37:30
But so for characters in general, large language models can
37:33
help to understand, especially with multimodal input.
37:38
Data can understand the world better, can do longer horizon
37:41
planning, and could basically help with our navigation tasks
37:45
because you need to somehow get from, you know, break down a larger
37:51
problem into tasks, subtasks, and then the navigation engine
37:56
could execute those subtasks.
37:57
For OLAF, we don't have any...
38:02
Any reasoning model, anytime soon, it's also, you know, you get
38:06
so much personality out if a human operator operates the... Thanks!
38:16
Is there one more question that I can answer?
38:25
Do you see in the future that character design will be based
38:29
more around manufacturability to bring those characters into the
38:33
parks as opposed to the character always coming first and then
38:39
the robot having to follow suit?
38:42
So manufacturability, basically optimizing robots
38:46
for manufacturability, that was the question.
38:49
So to some degree, if you've worked with modeler hardware components,
38:54
that already gets you a step there.
38:56
Because if you use production-grade actuators, production-grade
39:00
sensors, production-grade electronics inside, and their
39:04
modeler, you can use them across many characters.
39:08
That's basically already helping in the process.
39:11
But you probably think of components and
39:13
manufacturability of components.
39:15
And there are always a few refinement cycles that need
39:18
to happen to make a character truly production ready or
39:21
robot production ready.
39:23
So I think, you know, for us, manufacturability, we
39:27
don't do that at scale as much.
39:30
We customize characters.
39:31
We build only a few OLAFs.
39:33
You never will see, other than in a simulation video,
39:37
several OLAFs on a stage.
39:40
So we optimize basically for fewer characters but building
39:45
custom characters more quickly.
39:47
So manufacturability is an important item for us but
39:51
maybe for our use case not the most important one.
40:02
Shall we take another question?
40:06
We're technically at time, so we should say goodbye.
40:10
Sorry, my ears were icy.
40:12
Say that again?
40:14
Should say goodbye.
40:16
Goodbye, friends.
40:18
I'll never forget most of you.
40:23
So thanks so much for coming.
40:26
Absolutely.
40:29
Oh, thank you.
# Disney’s Olaf: From the Screen to Reality via Physical AI
Moritz Baecher,Director, Research Lab Zurich,Disney Research Imagineering
Rate Now
The translation of animated characters like Olaf, or the design of characters like our BDX Droids, requires balancing artistic fidelity with the constraints of the physical world. This talk explores how these robotic characters achieve that balance through the integration of modular mechatronics and deep reinforcement learning. Leveraging GPU-accelerated simulation to handle complex mechanisms, we train control policies that enable our characters to self-balance and self-regulate temperature. We further demonstrate progress in believable, autonomous navigation and human-robot interaction driven by onboard perception.
### Learn More About This Topic
Share
Favorite
Add to list
Events & Trainings:GTC San Jose
Date:March 2026
Industry:All Industries
Level:General Interest
Topic:Robotics - Robotics Simulation
Language:English
NVIDIA technology:CUDA, Isaac
Region:
Company Information
* [About Us](https://www.nvidia.com/en-us/about-nvidia/)
* [Investors](https://investor.nvidia.com/home/default.aspx)
* [Venture Capital (NVentures)](https://www.nvidia.com/en-us/startups/nventures/)
* [NVIDIA Foundation](https://www.nvidia.com/en-us/foundation/)
* [Research](https://www.nvidia.com/en-us/research/)
* [Corporate Sustainability](https://www.nvidia.com/en-us/sustainability/)
* [Technologies](https://www.nvidia.com/en-us/technologies/)
* [Careers](https://www.nvidia.com/en-us/about-nvidia/careers/)
News and Events
* [Newsroom](https://nvidianews.nvidia.com/)
* [Company Blog](https://blogs.nvidia.com/)
* [Technical Blog](https://developer.nvidia.com/blog/)
* [Webinars](https://www.nvidia.com/en-us/about-nvidia/webinar-portal/)
* [Stay Informed](https://www.nvidia.com/en-us/preferences/email-signup/)
* [Events Calendar](https://www.nvidia.com/en-us/events/)
* [GTC AI Conference](https://www.nvidia.com/gtc/events/)
* [NVIDIA On-Demand](https://www.nvidia.com/en-us/on-demand/)
Popular Links
* [Developers](https://developer.nvidia.com/)
* [Partners](https://www.nvidia.com/en-us/about-nvidia/partners/)
* [Executive Insights](https://www.nvidia.com/en-us/executive-insights/)
* [Startups and VCs](https://www.nvidia.com/en-us/startups/)
* [Documentation](https://docs.nvidia.com/)
* [Technical Training](https://www.nvidia.com/en-us/learn/organizations/)
* [Professional Services for Data Science](https://www.nvidia.com/en-us/support/enterprise/advisory-services/)
Follow NVIDIA
[](https://www.facebook.com/NVIDIA "")[](https://www.instagram.com/nvidia/?hl=en)[](https://www.linkedin.com/company/nvidia/)[](https://twitter.com/nvidia "")[](https://www.youtube.com/user/nvidia)
[United States](https://www.nvidia.com/en-us/location-selector/)
* [Privacy Policy](https://www.nvidia.com/en-us/about-nvidia/privacy-policy/)
* [Your Privacy Choices](https://www.nvidia.com/en-us/about-nvidia/privacy-center/)
* [Terms of Service](https://www.nvidia.com/en-us/about-nvidia/terms-of-service/)
* [Accessibility](https://www.nvidia.com/en-us/about-nvidia/accessibility/)
* [Corporate Policies](https://www.nvidia.com/en-us/about-nvidia/company-policies/)
* [Product Security](https://www.nvidia.com/en-us/product-security/)
* [Contact](https://www.nvidia.com/en-us/contact/)
Copyright © 2026 NVIDIA Corporation
Select Location
The Americas
* [Argentina](https://www.nvidia.com/es-la/ "Argentina")
* [Brasil (Brazil)](https://www.nvidia.com/pt-br/ "Brasil (Brazil)")
* [Canada](https://www.nvidia.com/en-us/ "Canada")
* [Chile](https://www.nvidia.com/es-la/ "Chile")
* [Colombia](https://www.nvidia.com/es-la/ "Colombia")
* [México (Mexico)](https://www.nvidia.com/es-la/ "México (Mexico)")
* [Peru](https://www.nvidia.com/es-la/ "Peru")
* [United States](https://www.nvidia.com/en-us/ "United States")
Europe
* [België (Belgium)](https://www.nvidia.com/nl-nl/ "België (Belgium)")
* [Belgique (Belgium)](https://www.nvidia.com/fr-be/ "Belgique (Belgium)")
* [Česká Republika (Czech Republic)](https://www.nvidia.com/cs-cz/ "Česká Republika (Czech Republic)")
* [Danmark (Denmark)](https://www.nvidia.com/da-dk/ "Danmark (Denmark)")
* [Deutschland (Germany)](https://www.nvidia.com/de-de/ "Deutschland (Germany)")
* [España (Spain)](https://www.nvidia.com/es-es/ "España (Spain)")
* [France](https://www.nvidia.com/fr-fr/ "France")
* [Italia (Italy)](https://www.nvidia.com/it-it/ "Italia (Italy)")
* [Nederland (Netherlands)](https://www.nvidia.com/nl-nl/ "Nederland (Netherlands)")
* [Norge (Norway)](https://www.nvidia.com/nb-no/ "Norge (Norway)")
* [Österreich (Austria)](https://www.nvidia.com/de-at/ "Österreich (Austria)")
* [Polska (Poland)](https://www.nvidia.com/pl-pl/ "Polska (Poland)")
* [România (Romania)](https://www.nvidia.com/ro-ro/ "România (Romania)")
* [Suomi (Finland)](https://www.nvidia.com/fi-fi/ "Suomi (Finland)")
* [Sverige (Sweden)](https://www.nvidia.com/sv-se/ "Sverige (Sweden)")
* [Türkiye (Turkey)](https://www.nvidia.com/tr-tr/ "Türkiye (Turkey)")
* [United Kingdom](https://www.nvidia.com/en-gb/ "United Kingdom")
* [Rest of Europe](https://www.nvidia.com/en-eu/ "Rest of Europe")
Asia
* [Australia](https://www.nvidia.com/en-au/ "Australia")
* [中国大陆 (Mainland China)](https://www.nvidia.com/zh-cn/ "中国大陆 (Mainland China)")
* [India](https://www.nvidia.com/en-in/ "India")
* [日本 (Japan)](https://www.nvidia.com/ja-jp/ "日本 (Japan)")
* [대한민국 (South Korea)](https://www.nvidia.com/ko-kr/ "대한민국 (South Korea)")
* [Singapore](https://www.nvidia.com/en-sg/ "Singapore")
* [台灣 (Taiwan)](https://www.nvidia.com/zh-tw/ "台灣 (Taiwan)")
Middle East
* [Middle East](https://www.nvidia.com/en-me/ "Middle East")
NVIDIA uses cookies to improve your experience on our web site. We and our third-party partners also use cookies and other tools to collect and record information you provide as well as information about your interactions with our websites for performance improvement, analytics, and to assist in marketing efforts. By clicking "Accept All", you consent to our use of cookies and other tools as described in our [Cookie Policy](https://www.nvidia.com/en-us/about-nvidia/cookie-policy/). You can manage your cookie settings by clicking on "Manage Settings." By continuing to use this site or by clicking one of the buttons below, you agree to our [Terms of Service](https://www.nvidia.com/en-us/about-nvidia/terms-of-service/) (which contains important waivers). Please see our [Privacy Policy](https://www.nvidia.com/en-us/about-nvidia/privacy-policy/) for more information on our privacy practices.
We have detected the Global Privacy Control (GPC) signal and recorded your rejection of all optional cookies on this site for this browser. You can manage your cookie settings by clicking on "Manage Settings". Please see our [Cookie Policy](https://www.nvidia.com/en-us/about-nvidia/cookie-policy/) for more information. To opt out of non-cookie personal information "sales" / "sharing" for targeted advertising purposes, please visit the [NVIDIA Preference Center](https://www.nvidia.com/en-us/about-nvidia/privacy-center/). Please see our [Privacy Policy](https://www.nvidia.com/en-us/about-nvidia/privacy-policy/) for more information on our privacy practices.
We have detected the Global Privacy Control Signal (GPC) and have opted you out of all optional cookies on this browser. You can manage your cookie settings by clicking on "Manage Settings". Please see our [Cookie Policy](https://www.nvidia.com/en-us/about-nvidia/cookie-policy/) for more information. We have also opted you out of "sharing"/"sales" of personal information outside of cookies. You can manage these settings in the NVIDIA [NVIDIA Preference Center](https://www.nvidia.com/en-us/privacy-center/). Please see our [Privacy Policy](https://www.nvidia.com/en-us/about-nvidia/privacy-policy/) for more information.
We have detected the Global Privacy Control Signal (GPC) and have opted you out of all optional cookies on this browser. You can manage your cookie settings by clicking on "Manage Settings". Please see our [Cookie Policy](https://www.nvidia.com/en-us/about-nvidia/cookie-policy/) for more information. We have also opted you out of "sharing"/"sales" of personal information outside of cookies which overrides at least one of your previous settings. You can manage them in the [NVIDIA Preference Center](https://www.nvidia.com/en-us/privacy-center/). Please see our [Privacy Policy](https://www.nvidia.com/en-us/about-nvidia/privacy-policy/) for more information.
Manage Settings
Reject Optional Accept All
Cookie Settings
We and our third-party partners (including social media, advertising, and analytics partners) use cookies and other tracking technologies to collect, store, monitor, and process certain information about you when you visit our website. The information collected might relate to you, your preferences, or your device. We use that information to make the site work, analyze performance and traffic on our website, provide a more personalized web experience, and assist in our marketing efforts.
Under certain privacy laws, you have the right to direct us not to "sell" or "share" your personal information for targeted advertising. To opt-out of the "sale" and "sharing" of personal information through cookies, you must opt-out of optional cookies using the toggles below. To opt out of the "sale" and "sharing" of data collected by other means (e.g., online forms) you must also update your data sharing preferences through the [NVIDIA Preference Center](https://www.nvidia.com/en-us/about-nvidia/privacy-center/).
Click on the different category headings below to find out more and change the settings according to your preference. You cannot opt out of Required Cookies as they are deployed to ensure the proper functioning of our website (such as prompting the cookie banner and remembering your settings, etc.). By clicking "Save and Accept" or "Decline All" at the bottom, you consent to the use of cookies and other tools as described in our [Cookie Policy](https://www.nvidia.com/en-us/about-nvidia/cookie-policy/) in accordance with your settings and accept our [Terms of Service](https://www.nvidia.com/en-us/about-nvidia/terms-of-service/) (which contains important waivers). For more information about our privacy practices, please see our [Privacy Policy](https://www.nvidia.com/en-us/about-nvidia/privacy-policy/).
Required Cookies
Always Active
These cookies enable core functionality such as security, network management, and accessibility. These cookies are required for the site to function and cannot be turned off.
Cookies Details
Performance Cookies
- [x] Performance Cookies
These cookies are used to provide quantitative measures of our website visitors, such as the number of times you visit, time on page, your mouse movements, scrolling, clicks and keystroke activity on the websites; other browsing, search, or product research behavior; and what brought you to our site. These cookies may store a unique ID so that our system will remember you when you return. Information collected with these cookies is used to measure and find ways to improve website performance.
Cookies Details
Personalization Cookies
- [x] Personalization Cookies
These cookies collect data about how you have interacted with our website to help us improve your web experience, such as which pages you have visited. These cookies may store a unique ID so that our system will remember you when you return. They may be set by us or by third party providers whose services we have added to our pages. These cookies enable us to provide enhanced website functionality and personalization as well as make the marketing messages we send to you more relevant to your interests. If you do not allow these cookies, then some or all of these services may not function properly.
Cookies Details
Advertising Cookies
- [x] Advertising Cookies
These cookies record your visit to our websites, the pages you have visited and the links you have followed to influence the advertisements that you see on other websites. These cookies and the information they collect may be managed by other companies, including our advertising partners, and may be used to build a profile of your interests and show you relevant advertising on other sites. We and our advertising partners will use this information to make our websites and the advertising displayed on it, more relevant to your interests.
Cookies Details
Cookie List
Clear
* - [x] checkbox label label
Apply Cancel
Consent Leg.Interest
- [x] checkbox label label
- [x] checkbox label label
- [x] checkbox label label
Decline All Save and Accept
[](https://www.onetrust.com/solutions/consent-and-preferences/)
Copy debug info