Stefan Wagner (@twbompo) header
Stefan Wagner (@twbompo) profile picture

Games developer at Mobile Heroes (Tiny Armies, Hi Frog!, ReBounce, Totem Tower) Develops games and apps with libGDX and Unity. @bompo@mastodon.gamedev.place

Dresden, Germany
Joined March 2009
Photo by Stefan Wagner (@twbompo)Photo by Stefan Wagner (@twbompo)Photo by Stefan Wagner (@twbompo)Photo by Stefan Wagner (@twbompo)Photo by Stefan Wagner (@twbompo)Photo by Stefan Wagner (@twbompo)Photo by Stefan Wagner (@twbompo)Photo by Stefan Wagner (@twbompo)Photo by Stefan Wagner (@twbompo)Photo by Stefan Wagner (@twbompo)
Stefan Wagner retweeted
(1/5) Everyone says dexterous manipulation needs real-world data. These policies were trained with none. Our robots mesh gears, thread nuts, and insert pegs zero-shot from RGB, and the same recipe works for locomotion across terrains much harder than before. The missing piece wasn't a better RL algorithm, but a small tweak to what the robot practices. Today we're releasing Success-Guided Sampling (SGS), a simple change to the task sampler that unlocks new levels of dexterity and agility with large-scale, end-to-end sim2real RL. No demonstrations, no per-task reward tuning. Just PPO at scale. Check out the policies yourself at sgs-rl.github.io! All videos at 1x speed! 🧵 🔈
43
185
70
1,082
214,075
Stefan Wagner retweeted
(1/6) From @GoogleDeepmind’s AlphaGo to AlphaZero, we saw machines find moves that reached beyond what we’d imagined for them. No one taught them those strategies. They emerged from reinforcement learning: playing, failing, and trying again under a single binary task reward for success or failure. Inspired by them, we couldn’t help but wonder: what might a robot discover if we gave it that same freedom to learn? That question led us to AlphaNIST. Starting tabula rasa, with that same binary reward plus only a simple control cost, a single policy solves all 20 @NIST assembly tasks at a 98.2% success rate in simulation. More surprising still, AlphaNIST can complete 10 different tasks in a row, receiving that reward only once, after all 10 are done. Just as AlphaGo’s strategies surprised its creators, AlphaNIST discovered dexterous strategies we couldn’t have come up with ourselves. Paper: arxiv.org/abs/2610.13354 Website: alpha-nist.github.io
64
338
124
2,688
425,479
Stefan Wagner retweeted
Here's Qwen 3.6 27B on an AMD 7900XTX over USB3 (literally any computer made in the last 10 years) at 34 tok/s. The eGPU dock that supports this launches on the 12th, with 100% open source firmware and an extra USB port for serial + unbrickability.
65
101
23
1,929
114,826
Stefan Wagner retweeted
Been building a system that turns point clouds and scans into reconstructed, editable environments. The goal isn’t just visual fidelity. Walls, floors, ceilings, and objects should remain separate so users and agents can understand and modify the scene.
17
33
1
479
123,222
Stefan Wagner retweeted
I’m happy to announce the release of a new open source 3D physics engine called Box3D. I’ve been working on this project for a few years now, but it represents over 20 years of experience writing physics engines for games. Read more here: box2d.org/posts/2026/06/anno…
109
760
69
5,785
1,073,650
Stefan Wagner retweeted
📢📢GenRecon: Bridging Generative Priors for Multi-View 3D Scene Reconstruction📢📢 Reconstructing high-fidelity 3D scenes from sparse RGB input is hard. It needs a strong 3D prior! We reformulate multi-view scene reconstruction as conditional 3D generation over overlapping spatial chunks, lifting posed image features into a generative shape prior via 3D conditioning. As an example prior, we build on Trellis2, and train it such that its reconstruction is pixel aligned and matches from all views. GenRecon achieves unprecedented reconstruction quality from any sparse RGB input sequence, even from a phone capture. The reconstruction also includes PBR materials which facilitates relighting and virtual object insertion. youtu.be/Tp-i06DPXa0 kasothaphie.github.io/GenRec… Amazing work by @katha_schmid, @nicolasvluetzow, Jozef, @angelaqdai
7
62
2
311
18,508
Stefan Wagner retweeted
We’ve released the official Blender add-on for ZOZO’s Contact Solver, our open-source physics solver developed at ZOZO, Inc. Blender users, please take a look! github.com/st-tech/ppf-conta…
55
949
64
8,079
608,882
Stefan Wagner retweeted
LLM post-training used to mean fine-tuning to a downstream task Robotics has been stuck in this setting, needing task-specific fine-tuning for best performance π07 changes this: It works out of the box & outperforms fine-tuned specialists Details: pi.website/pi07
32
67
13
638
83,253
Stefan Wagner retweeted
LeRobot is doing important work for robotics by making datasets, models, and robots more accessible! Check out these policies
Releasing the Unfolding Robotics blog! Time to unfold robotics: we trained a robot to fold clothes using 8 bimanual setups, 100+ hours of demonstrations, and 5k+ GPU hours. Flashy robot demos are everywhere. But you rarely see the real story: the data, the failures, the engineering. We’re sharing everything: code, data, and details in the blog → huggingface.co/spaces/lerobo…
5
16
117
11,850
Stefan Wagner retweeted
Putting my tokens where my mouth is. I built pi-share-hf. Share your pi coding agent sessions as @huggingface datasets. github.com/badlogic/pi-share… It tries to prevent you from uploading sessions containing PII/sensitive data with 3 tiers of defenses. Best used on OSS coding sessions, as those are less likely to contain sensitive info. Uses pi agents for PII detection, which can cost you a lot of tokens. Read the README with your human eyes so you don't accidentally pwn yourself or get a huge bill. Haven't figured out how to filter for such datasets on HF yet. @ClementDelangue, any pointers on how to best label them so people can find them?
we as software engineers are becoming beholden to a handful of well funded corportations. while they are our "friends" now, that may change due to incentives. i'm very uncomfortable with that. i believe we need to band together as a community and create a public, free to use repository of real-world (coding) agent sessions/traces. I want small labs, startups, and tinkerers to have access to the same data the big folks currently gobble up from all of us. So we, as a community, can do what e.g. Cursor does below, and take back a little bit of control again. Who's with me? cursor.com/blog/real-time-rl…
17
47
7
374
52,160
Stefan Wagner retweeted
A glimpse of the absolute insanity in numbers that is our industry right now. Attributed Claude activity over the last 90 days on GitHub. The pushers did us dirty over Christmas.
22
38
22
431
230,057
Stefan Wagner retweeted
I'm usually not one to write thought pieces without much technical depth. But here we go. Slow the fuck down. mariozechner.at/posts/2026-0…
150
493
176
2,865
630,070
Super excited to announce FlexAvatar! 📢📢 With FlexAvatar, you can create a full 360°, high-quality, and expressive 3D head avatar from just a single portrait image. In this real-time demo, we showcase the avatar creation which takes only 2 minutes. 👉tobias-kirschstein.github.io…
Want to create an avatar from a single image? FlexAvatar is a transformer model that creates full 360°, high-quality, and expressive 3D head avatar from just a single portrait image in minutes. Real-time Demo: FlexAvatar's lightweight architecture allows both animation and rendering in real-time, enabling interactive user experiences. To create a new 3D head avatar, only one image is required, e.g., from a webcam. The final avatar is ready after 2 minutes. Architecture: Under the hood, FlexAvatar adopts a transformer-based encoder-decoder design. The encoder maps the input image onto a latent avatar space, while the decoder produces 3D Gaussian attribute maps by incorporating the animation signal via cross-attention. The model learns all facial animations directly from the data without relying on pre-built 3D face models. This equips the avatars with realistic facial expressions. The internal avatar latent space can be conveniently used to integrate additional observations of a person via fitting. This enables use-cases where more than one image of a person is available, e.g., from a phone scan of the person. We train jointly on 2D monocular videos and multi-view data. However, in monocular videos, the animation signal leaks the target viewpoint, causing the model to produce incomplete 3D heads. We call this phenomenon entanglement of driving signal and target viewpoint. To prevent entanglement, we introduce bias sinks. These are learnable tokens that indicate whether a training sample stems from a monocular or a multi-view dataset. During training, the model learns to produce incomplete 3D heads only when the monocular token is present. During inference, FlexAvatar then always uses the multi-view token for which the model has learned to produce complete 3D heads. This simple design allows to combine the generalizability from monocular data with the quality of multi-view data. FlexAvatar summary: - Input: Single-image, phone scan, or monocular video - Output: Full 360° head avatar - Expressive animations - Real-time rendering and animation - Generalization to any portrait - Create a new avatar in 2 minutes - Use bias sinks to combine 2D and 3D data 🏠tobias-kirschstein.github.io… 🌍arxiv.org/pdf/2512.15599 🎥youtu.be/g8wxqYBlRGY Great work by @TobiasKirschst1 and @SGiebenhain!
13
65
4
486
47,477
Stefan Wagner retweeted
finally published this visual article on dithering (part 1)! I had so much fun and learned a lot while making this. not only about dithering, but also about #threejs shaders. I hope you do too! as usual, animated with #animejs enjoy! 🔗 visualrambling.space/ditheri…
74
162
42
1,758
117,375
wave_tracer 0.1 released. wavetracer.dev wave_tracer combines path tracing and wave optics in a novel way, for practical general-purpose wave simulations across a variety of EM modalities and applications. This is an early alpha release.
49
2
200
16,877
I've written up how I approach "agentic engineering" for real-world codebases & tasks. TL;DR: Treat LLMs like shitty computers you program with natural language, not chatbots you beg for code. mariozechner.at/posts/2025-0…
11
34
9
233
35,407
Stefan Wagner retweeted
Great news for the entire Blender community! I am happy to announce that I received an Epic Mega Grant to finish integration of SLIM into the Blender official codebase (so that it is free for everyone). And now the task is done: uvpackmaster.com/slim/ #b3d #Blender3d #3dcg
31
303
29
2,239
192,930
What structural task representation enables multi-stage, in-the-wild, bimanual, reactive manipulation? Introducing ReKep: LVM to label keypoints & VLM to write keypoint-based constraints, solve w/ optimization for diverse tasks, w/o task-specific training or env models. 🧵👇
18
102
20
515
192,030
One line, one word. ✏️🔤 A simple #p5js sketch where you draw a line and Gemini tries to describe it in a single word. I find it really fun to have the model write on the canvas with me (wherever drawing finishes). Uses Gemini 1.5 Flash API. Try it + fork/remix my sketch here: editor.p5js.org/alexanderche…
13
62
11
448
72,380
Stefan Wagner retweeted
- full end-to-end policy - no teleoperation - quasi-full open-source software and hardware (even Reachy-1’s hardware is open-source CC-BY-SA CAO assembly)
19
69
9
390
111,317