Nvidia GTC conference AI Conference & Expo March 16–19, 2026

[Anthony Andrade]

Hello Homebrew Robotics Club team,

I hope you’re doing well. I have been following the Homebrew Robotics Club for a while now—really appreciate the kind of community you’ve built around robotics.

I wanted to ask for your perspective on an upcoming instructor-led workshop which is $495:

“How to Simulate, Train, Validate, and Deploy an End-to-End Robotics Workflow with NVIDIA Isaac”
(Sunday, March 15, 9:00 a.m.–5:00 p.m. PDT)

The workshop focuses on using Isaac Sim and Isaac Lab to train robot policies in simulation, validate them through software-in-the-loop testing, and deploy them to real hardware, with emphasis on sim-to-real techniques.

I’m currently working on a personal humanoid robotics project and already use tools like Python, Linux, URDFs, and simulation environments. My main goal is to deepen my understanding of end-to-end workflows—especially policy training, validation, and deployment—rather than just following canned demos.

From your experience, would you say this type of workshop is genuinely valuable for someone building and iterating on their own robot hardware? Or is the material closer to a high-level overview that might be redundant for someone already experimenting with Isaac Sim / Lab on their own?

I’d really appreciate any insight you’re willing to share. Thank you for everything you do to support the local robotics community.

Best regards,
Anthony Andrade

[Stephen Williams]

Because I already registered for virtual GTC, I can't actually get to registration for that training now, with my main email address anyway.

I am interested in this area also.  The way I would like to approach this is to try to find time to work through the online documentation for Isaac.  Then I should have a much better idea of whether paying for that training is worth it.

Looks like there is a lot here: https://developer.nvidia.com/search?q=isaac%20robot%20training&page=1

Sometimes those kinds of workshops are good for jump starting into an area.  Other times, they are good for accelerating when you already have some experience and are partially up to speed.  This offering should be the former, but I can't tell yet.  Google brings up this video when searching for that training:

https://www.youtube.com/watch?v=tQziqSx-F80&t=2209s

Stephen

--
You received this message because you are subscribed to the Google Groups "HomeBrew Robotics Club" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hbrobotics+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/hbrobotics/59fc6d72-9d2e-4b6a-8f39-a1f1347e6db8n%40googlegroups.com.

--

Stephen D. Williams Founder: VolksDroid, Blue Scholar Foundation 650-450-8649 | fax:703-995-0407 | s...@lg.net | https://VolksDroid.org | https://BlueScholar.org | https://sdw.st/in

[Chris Albertson]

The Nvidia content is VERY Nvidia-centric and light on theory.  It is more of “how to use our product” than “how AI works”.   

If you are a company trying to make money, the “how to use our product” kind of training is what you want to produce.  Especially because the more theoretical stuff has some prerequisites that seriously reduce the size of your audience.

It depends on what you want.   Do you need to get the job done, and you have already decided to go all-in with Nvieda?   Then spend or a well-spec’d Orin and the training.  But if you want to understand this tech, you might want a broader base

Here is a way to think about it.    A few days ago, Apple released a software upgrade that allows “RDMA of Thunderbolt 5”.     What this means is that two or more Macs can now access each other’s RAM over an 80 Gb/second “rat’s nest” topology network, and the network is just cables and existing ports.  So each computer has three ports of 80 Gbps each, then you have 240 Gbps connecting a stack of computers.      Now compare the price of a stack of Apple Mac Studio computers with a total of 1TB RAM and Nvidia’s solution.   Apple is like 10X cheaper for the same performance.     As a hobbyist, do you care about knowing enough to evaluate a solution such as Apple’s or do you want to get locked into one vendor?   This is a real question because going all-in with Nvidia is not bad.  You get stuff done.

[Jeremy Williams]

Well said Chris.

As for course or not, I’m not a big fan of Nvidia and I know many in the field echo that sentiment. 

Too many times they have decided to switch direction and stranded ‘high and dry’ previous product lines. Doesn’t mean they don’t have impressive products, but I would never go full in on anything they have. For a short term experiment or to cobble together something for a presentation or proof of concept - sure. Great. 

Otherwise I’d not invest too much time into their biosphere. Learn or take what you need right now to do what you need to do. 

And honestly, they have material to teach that’s free. And if it’s not free or too hard to get, that’s maybe a sign that their level of support is lacking and it may not be ready for prime time. 

Just my two cents.

Good luck:)

Jeremy Williams 

--
You received this message because you are subscribed to the Google Groups "HomeBrew Robotics Club" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hbrobotics+...@googlegroups.com.

To view this discussion visit https://groups.google.com/d/msgid/hbrobotics/EE795680-6AF0-4D38-9EAB-F0C25350D89C%40gmail.com.

[s...@lig.net]

I agree mostly with both Jeremy & Chris. I want to go deep in certain areas, to understand most of the details about how things work, and to be able to experiment with variations, remixes, or new ideas.

However, just getting into a mode where you can work on a particular area can be tough. One area from NVidia that seems really valuable is having high quality physics-based virtual environments with environment generation (now with generative AI) along with various algorithms, components, and infrastructure that is ready to go. Building any of that is a lifetime pursuit by itself. Working within that, lowering friction to get to an interesting problem space, seems very useful. Apple isn't helping with that. There are open source options, ROS etc., that get you there to some extent. But at least having the commercial option, that seems free for a wide range of use on the fastest hardware, is valuable.

Having RDMA with Mac Studio units is huge. The capability has been in the hardware for a while. Impressive & useful that they finally wired it in as a standard thing. The Mac Studios are powerful, can go up to 512GB of fast unified RAM (but not yet with the newest CPU/GPUs), and have some reasonable GPU capability. However, except for size of RAM, they are still far behind the performance of something like an RTX Pro 6000. For some cases, size of memory is more important, and the distributed compute can be competitive. But for other work, the super fast VRAM + massively parallel compute + special compute modules like NVENC video codecs, mean that nothing really competes with NVidia right now for the full range of AI / ML / simulation / video & other related capabilities.

Hopefully other options will keep up. But it seems they likely plateau deliberately because it is more reasonable to go after useful, but not super hard targets. Just keeping a reasonable power + heat envelope rules much of that out.

So, what is best depends on what you want to do. I decided that for development & exploration, something other than NVidia would constantly feel that I was limited & missing out. Now, that is less true with EXO Labs and others doing great work. There was an interesting benchmark that paired a number of Mac Studio systems with NVidia GPX Spark units, over 10Gbit ethernet which you can get from Thunderbolt also. The resulting system performed far better than either type of hardware alone. EXO was able to use the best of both for appropriate stages, in parallel and pipelined. That is going to be the future.

But that still doesn't get you the super broad, powerful, integrated simulation environment that is very useful for RL training. There are open source simulators. If they keep up, or even surpass NVidia, that would be great. But I won't wait for that.

sdw

> >> *“How to Simulate, Train, Validate, and Deploy an End-to-End
> >> Robotics Workflow with NVIDIA Isaac”*

> >> (Sunday, March 15, 9:00 a.m.–5:00 p.m. PDT)
> >>
> >> The workshop focuses on using Isaac Sim and Isaac Lab to train
> >> robot policies in simulation, validate them through software-in-
> >> the-loop testing, and deploy them to real hardware, with emphasis
> >> on sim-to-real techniques.
> >>
> >> I’m currently working on a personal humanoid robotics project and
> >> already use tools like Python, Linux, URDFs, and simulation
> >> environments. My main goal is to deepen my understanding of end-
> >> to-end workflows—especially policy training, validation, and
> >> deployment—rather than just following canned demos.
> >>
> >> From your experience, would you say this type of workshop is
> >> genuinely valuable for someone building and iterating on their
> >> own robot hardware? Or is the material closer to a high-level
> >> overview that might be redundant for someone already
> >> experimenting with Isaac Sim / Lab on their own?
> >>
> >> I’d really appreciate any insight you’re willing to share. Thank
> >> you for everything you do to support the local robotics community.
> >>
>
> --
> You received this message because you are subscribed to the Google
> Groups "HomeBrew Robotics Club" group.
> To unsubscribe from this group and stop receiving emails from it,
> send an email to hbrobotics+...@googlegroups.com

> <mailto:hbrobotics+...@googlegroups.com>.

> To view this discussion visit https://groups.google.com/d/msgid/
> hbrobotics/EE795680-6AF0-4D38-9EAB-F0C25350D89C%40gmail.com

> <https://groups.google.com/d/msgid/hbrobotics/
> EE795680-6AF0-4D38-9EAB-F0C25350D89C%40gmail.com?
> utm_medium=email&utm_source=footer>.

>
> --
> You received this message because you are subscribed to the Google
> Groups "HomeBrew Robotics Club" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to hbrobotics+...@googlegroups.com

> <mailto:hbrobotics+...@googlegroups.com>.

> To view this discussion visit https://groups.google.com/d/msgid/

> hbrobotics/CAJHxCZG84sxrV6C7DmW6Cso_r7dg02d9qx8X2-
> X6imK0ZF1RXw%40mail.gmail.com <https://groups.google.com/d/msgid/
> hbrobotics/CAJHxCZG84sxrV6C7DmW6Cso_r7dg02d9qx8X2-
> X6imK0ZF1RXw%40mail.gmail.com?utm_medium=email&utm_source=footer>.
>

[Stephen Williams]

I didn't get to the NVidia area, and I see now that they ended early yesterday.  Some of the talks seem interesting.  Hopefully they will be online eventually.

https://www.nvidia.com/en-us/events/ces/

https://www.youtube.com/@NVIDIA

Stephen