Clemens Marschner

4.9K posts

Clemens Marschner banner
Clemens Marschner

Clemens Marschner

@cmarschner

Principal Engineer @robco_robotics. Building next generation, AI-based robots. Ex @lyft, Microsoft. Work with us:

Munich Katılım Mart 2008
998 Takip Edilen990 Takipçiler
Sabitlenmiş Tweet
Clemens Marschner
Clemens Marschner@cmarschner·
Most autonomy stacks never make it into real factories. Ours do. We’re bringing Physical AI into production in Munich - and we’re hiring for our Autonomy Team. See below.
RobCo@Robco_Robotics

We’re hiring for our Autonomy Team in Munich 🚀 Backed by our $100M Series C, we bring Physical AI into real production Open roles in Robotics, AI, 3D Perception, Mechanical & Mechatronics Engineering. Build autonomy that runs in factories today.

English
2
2
15
1.2K
Clemens Marschner
Clemens Marschner@cmarschner·
One of the hardest parts when putting out open positions these days is the bots that spam you for the first two weeks. Most of these applicants are not qualified. But they consume a lot of resources if you want to treat everyone fairly. It forces the employers now to strike back and use AI tools to filter them out. Not sure if this is net positive.
Tips Excel@gudanglifehack

Claude Cowork can apply to 50 jobs in under 30 minutes. Here's how to set it up.

English
0
0
0
69
Clemens Marschner
Clemens Marschner@cmarschner·
If you are at Hannover Messe, join us in Hall 26!
Lukas Ziegler@lukas_m_ziegler

If you’re at Hannover Messe, @Robco_Robotics is showcasing its systems in Hall 26, Booth E35. One of the highlights is a DIY modularity station, where visitors can build and reconfigure a robot setup using real components. The idea is to demonstrate how modular robotics systems adapt to the use case, not the other way around They’re also showing how their digital twin infrastructure supports faster deployment and reconfiguration. Another piece of the stack is RobFlow, their no-code interface for programming and operating robots. The concept is straightforward: if you can operate a tablet, you should be able to configure an automation workflow. RobCo is also giving a preview of its next generation autonomous product line. The RobCo team, from founders to engineers and integration specialists, will be on site throughout the event. If you’re exploring practical industrial automation, Booth E35 in Hall 26 is worth a stop at HANNOVER MESSE. I’ll be there too, and hope to see you there! 👋🏻 ~~ ♻️ Join the weekly robotics newsletter, and never miss any news → ziegler.substack.com

English
0
0
1
57
Jeffrey Emanuel
Jeffrey Emanuel@doodlestein·
I’m surprised that you don’t see OpenAI announcing major partnerships with FANUC and KUKA. These companies make extremely accurate and powerful robot arms that can do small-scale work all the way up to lifting and spinning entire car frames. The problem with them has always been that the programming is extremely tedious, finicky, and time consuming. So the cost of the equipment is just the beginning since you need integrator consultants to program them for you. And then they’re not flexible. Even after you finally dial everything in, you’re now stuck with that configuration. Good luck trying to optimize the overall layout of your plant or to change other work streams without needing to reprogram everything. But this is precisely where advanced AI, vision, RL, world models, etc. are so valuable. But it’s going to be hard for the big labs to match 30+ years of mechanical engineering excellence and learnings like the robot arm companies have. Not to mention 30+ year supplier relationships with every large Western auto OEM. And the arm companies certainly aren’t going to be training any frontier transformers soon! Anyway, that’s what I would be advising them to do now.
Jeffrey Emanuel tweet mediaJeffrey Emanuel tweet media
English
9
2
43
3.7K
Clemens Marschner retweetledi
Nainsi Dwivedi
Nainsi Dwivedi@NainsiDwiv50980·
Holy shit. The guy who BUILT Claude Code just shared his actual workflow. Boris Cherny runs 10-15 Claude sessions in parallel every single day. While you're prompting one AI, he has 5 in his terminal + 5-10 on the web all shipping code simultaneously. And the real weapon? His CLAUDE.md file. Every time Claude makes a mistake, the team adds a rule so it NEVER happens again. Boris literally said: "After every correction, end with: Update your CLAUDE.md so you don't make that mistake again." Claude writes rules for itself. The longer you use it, the smarter it gets on YOUR codebase. His other insane detail: he hasn't written a single line of SQL in 6+ months. Claude just pulls BigQuery data directly via CLI. Claude Code now accounts for 4% of ALL public GitHub commits. Engineers who haven't set this up yet are already behind. This CLAUDE.md template is the difference between using AI as a chatbot vs using it as a fleet of senior engineers. Drop it in any project. Free.
Nainsi Dwivedi tweet media
English
83
182
1.4K
354.1K
Clemens Marschner retweetledi
Yulu Gan
Yulu Gan@yule_gan·
Simply adding Gaussian noise to LLMs (one step—no iterations, no learning rate, no gradients) and ensembling them can achieve performance comparable to or even better than standard GRPO/PPO on math reasoning, coding, writing, and chemistry tasks. We call this algorithm RandOpt. To verify that this is not limited to specific models, we tested it on Qwen, Llama, OLMo3, and VLMs. What's behind this? We find that in the Gaussian search neighborhood around pretrained LLMs, diverse task experts are densely distributed — a regime we term Neural Thickets. Paper: arxiv.org/pdf/2603.12228 Code: github.com/sunrainyg/Rand… Website: thickets.mit.edu
Yulu Gan tweet media
English
88
433
3K
676.1K
Clemens Marschner
Clemens Marschner@cmarschner·
“As engineers develop with Claude Code, they’re seeing the friction to creating a new feature [decrease], and they’re seeing a much higher demand for code review.” While it’s fun to write code in hours that used to take days or weeks, the industry is now dependent on it - some money goes to NVidia, some to cloud providers, some to utilities, some to Anthropic. Across the whole software industry. Quite a shift. techcrunch.com/2026/03/09/ant…
English
0
0
1
110
Clemens Marschner
Clemens Marschner@cmarschner·
@jonstephens85 Do you think simulating on pixel level is necessary? Or is embedding space be the way to go.
English
0
0
0
38
Jonathan Stephens
Jonathan Stephens@jonstephens85·
This is one of the few sessions at GTC I cannot miss! Cosmos has been critical for speeding up the sim2real gap for autonomous vehicles and robotics. Record only one scene, and make multitudes of new scenes and scenarios. The world is infinitely complex and we are going to need these tools to simulate the long tail of scene variations. Cosmos is also one of my top priorities in 2026. Come find me at the conference if you want to talk world models. #GenerativeAI #WorldModel #Cosmos
NVIDIA DRIVE@NVIDIADRIVE

World models are transforming AV development by capturing both the environment and the world’s response with high fidelity. Join Sanja Fidler, VP of AI Research at NVIDIA and Associate Professor at the University of Toronto, to explore how NVIDIA Cosmos powers the real-time simulation and closed-loop evaluation that empowers developers to iterate faster and bring safer autonomous systems to the real world. 📅 Thursday, March 19 | 10:00 a.m. - 10:40 a.m. 📍 #NVIDIAGTC 2026 Add to schedule 👉 nvda.ws/4kRYo9s

English
2
8
36
4.6K
Clemens Marschner
Clemens Marschner@cmarschner·
The funny thing about Kung Fu-dancing robots is: We could do it in Europe or other places, we just choose not to. We don't care. It's all published work. The cool thing is actually a low-cost humanoid that can execute it.
Clemens Marschner tweet media
English
0
0
2
66
Jannik Grothusen
Jannik Grothusen@JannikGrothusen·
New control stack for TRLC-DK1: Joint-space PD + gravity compensation Before vs. after:
English
15
43
524
29.8K
Clemens Marschner
Clemens Marschner@cmarschner·
Glad I could speak about the importance of Physical AI in the European AI landscape today. Thanks @SPRIND for getting in touch! If you work in AI, here‘s your opportunity to get non-diluted start funding
Ilir Aliu@IlirAliu_

€125M in funding. Non-dilutive for 10 teams. Only three survive to build €1 billion AI labs. I know… Some people will say it’s too little. Others will say it’s too late. But when Alessandro and Mirko Holzer reached out about supporting this initiative by the @SPRIND, it was clear to me: as someone working closely with AI and robotics companies in Europe, I want to be part of this. - Next Frontier AI - is one of Europe’s most ambitious attempts to build independent frontier AI capabilities. This is not just a grant program. It is a long-term effort to create new AI labs in Europe that can compete globally. Munich kicks things off tonight. Next stop: Paris on March 19. Funding €125M non-dilutive funding for 10 teams Goal: create up to 3 frontier AI labs capable of raising €1B rounds Who can apply Individuals, small teams, or established groups with deep technical expertise in AI, robotics, systems, or large-scale computing Timeline Applications open May 2026 10 teams selected in July 2026 Program runs for 24 months Final winners announced Fall 2028 Outcome The goal is not just research. Teams are expected to build venture-ready AI labs with real infrastructure, pilots, teams, and investor-ready data rooms. This you? 🇫🇷 March 19, 2026 Registration open: luma.com/nfai-paris Was nice seeing you, @DominiqueCAPaul 👋

English
0
0
1
113
Michaela Kuefner
Michaela Kuefner@MKuefner·
China shows off its AI robotics edge. German Chancellor Merz gets a taste of how the roles between China and Germany reverse. Engineers partly uncomfortable with us filming. An attempt to shut us in a waiting room during the Chancellor’s visit failed. German press too unruly.
English
278
656
4.1K
778.4K
Clemens Marschner
Clemens Marschner@cmarschner·
Jensen has become a source of wisdom.
Founder Mode@Founder_Mode_

Jensen Huang: "People with really high expectations have very low resilience." "I think one of my great advantages is that I have very low expectations. And I mean that. Most of the Stanford graduates have very high expectations. And you deserve to have high expectations because you came from a great school. You were very successful. You're top of your class. Obviously, you were able to pay for tuition. And then you're graduating from one of the finest institutions on the planet. You're surrounded by other kids that are just incredible. You naturally have very high expectations. People with very high expectations have very low resilience. And unfortunately, resilience matters in success. I don't know how to teach it to you except for I hope suffering happens to you. And I was fortunate that I grew up with my parents providing a condition for us to be successful on the one hand, but there were plenty of opportunities for setbacks and suffering. And to this day, I use the phrase pain and suffering inside our company with great glee. And I mean that. Boy, this is going to cause a lot of pain and suffering. And I mean that in a happy way, because you want to train, you want to refine the character of your company. You want greatness out of them. And greatness is not intelligence. Greatness comes from character, and character isn't formed out of smart people. It's formed out of people who suffered. And so if I could wish upon you, I don't know how to do it. For all of you Stanford students, I wish upon you ample doses of pain and suffering."

English
0
0
2
219
Clemens Marschner
Clemens Marschner@cmarschner·
"Unlike proposals in some countries that might lean toward mandatory government ID linkage or centralized digital IDs, Australia's law explicitly prohibits platforms from forcing users to use government-issued ID or an accredited Australian Government digital ID service as the only method for age proof. Platforms must offer reasonable alternatives to avoid compulsory government ID use. Common methods platforms use (or are expected to use) include: Age estimation / inference — AI-based facial age estimation (e.g., via selfie or video analysis), behavioral analysis (online patterns, device history), or statistical inference. Platforms often get only a "yes/no" (over/under 16) result, without retaining full biometric data. Age verification with documents — Scanning government ID (passport, driver's license), but only as an optional path — not mandatory. Other options — Bank-linked verification (e.g., via services like ConnectID in Australia), third-party providers that handle checks privately." (Grok)
English
1
0
0
31
Clemens Marschner
Clemens Marschner@cmarschner·
Social Media Algorithmen ("for you") sind auf dieselbe Weise süchtig machend wie Zigaretten und andere Drogen. Alternative ist daher, derart engagement-getriebene Seiten für alle Altersgruppen zu verbieten. Die Debatte, wie Alter festgestellt wird, wird in Deutschland geführt (Amthor-Bereich). Australien ist da schon weiter, und Spanien scheint das auch nicht zu stören. Letztlich wird sich der Jugendschutz durchsetzen. Pseudonymität wäre eine Variante. Spezielle anonyme Stellen eine andere.
Deutsch
1
0
0
30

Keşfet

@doodlestein @jonstephens85 @JannikGrothusen @SPRIND @chico_morazan @KayaSerhan_ @kurva_z_Hradce @MKuefner