Wen-Han Hsieh (@henseoba) header
Wen-Han Hsieh (@henseoba) profile picture
Pittsburgh, PA
Joined May 2023
Photo by Wen-Han Hsieh (@henseoba)Photo by Wen-Han Hsieh (@henseoba)Photo by Wen-Han Hsieh (@henseoba)Photo by Wen-Han Hsieh (@henseoba)Photo by Wen-Han Hsieh (@henseoba)
Really impressive to see how fast Genesis is growing! I interned there this summer and genuinely loved the experience. Super talented team and some of the nicest people I’ve worked with. Excited to see what’s next! 🚀
Genesis is entering its next phase. Today we've raised a $500M Series A at $3.5B. We can imagine far more than we can build physically today. Almost everything we are creating is still paid for in human time, by human hands. We want to remove the physical limits on human ambition. We are building general-purpose robots with human level intelligence and capability. Ones that work reliably across entire workflows for years. There are countless versions of possible futures where billions of robots are deployed everywhere, and we aspire to bring the brightest one that carries our hopes into reality.
16
465
Wen-Han Hsieh retweeted
What attracted me to Genesis from day one was the team and our shared ambition. Incredibly proud of what we've built together, and even more excited about what's ahead. We're hiring! If you're interested in joining us, feel free to reach out.
Genesis is entering its next phase. Today we've raised a $500M Series A at $3.5B. We can imagine far more than we can build physically today. Almost everything we are creating is still paid for in human time, by human hands. We want to remove the physical limits on human ambition. We are building general-purpose robots with human level intelligence and capability. Ones that work reliably across entire workflows for years. There are countless versions of possible futures where billions of robots are deployed everywhere, and we aspire to bring the brightest one that carries our hopes into reality.
3
4
56
2,786
Wen-Han Hsieh retweeted
So excited about our Series A and what it can enable us to achieve! I decided to join Genesis, because full-stack robotics has been its core. In the past four months, I've had the chance to work with world-class teams in learning, robotics, control, simulation, design, and more. It has been a terrific journey to see how we are pushing forward intelligence, dexterity, and everything else at scale. We are also hiring! If you are interested in building and shipping robots to the world to do stuff and shape the future, come and join us at genesis.ai/careers
Genesis is entering its next phase. Today we've raised a $500M Series A at $3.5B. We can imagine far more than we can build physically today. Almost everything we are creating is still paid for in human time, by human hands. We want to remove the physical limits on human ambition. We are building general-purpose robots with human level intelligence and capability. Ones that work reliably across entire workflows for years. There are countless versions of possible futures where billions of robots are deployed everywhere, and we aspire to bring the brightest one that carries our hopes into reality.
3
2
54
2,461
Wen-Han Hsieh retweeted
I've always been someone who follows my curiosity. This new funding gives us the opportunity to keep scaling our research, pursue even more ambitious ideas, and bring what we're building into the world. None of this would have been possible without the hard work of so many amazing people on our team. On a more personal note, I’m deeply grateful to the friends and mentors who have supported and influenced me along the way. I’ve been very lucky to meet so many kind and wise people early in my career. The future worth building is human. Onwards!
Genesis is entering its next phase. Today we've raised a $500M Series A at $3.5B. We can imagine far more than we can build physically today. Almost everything we are creating is still paid for in human time, by human hands. We want to remove the physical limits on human ambition. We are building general-purpose robots with human level intelligence and capability. Ones that work reliably across entire workflows for years. There are countless versions of possible futures where billions of robots are deployed everywhere, and we aspire to bring the brightest one that carries our hopes into reality.
1
4
40
2,226
Wen-Han Hsieh retweeted
Genesis is entering its next phase. Today we've raised a $500M Series A at $3.5B. We can imagine far more than we can build physically today. Almost everything we are creating is still paid for in human time, by human hands. We want to remove the physical limits on human ambition. We are building general-purpose robots with human level intelligence and capability. Ones that work reliably across entire workflows for years. There are countless versions of possible futures where billions of robots are deployed everywhere, and we aspire to bring the brightest one that carries our hopes into reality.
227
190
177
2,218
2,859,728
Wen-Han Hsieh retweeted
Anthropic acaba de lanzar el mejor truco para Claude Code: Fable 5.2 piensa → Sonnet 5.5 construye → Opus 5.5 revisa → Haiku 5.5 coordina un enjambre de agentes 1 minuto + el comando /advisor = 3,4 veces más barato Así se configura: Primero, entiende cómo funciona: • Esta configuración sirve para cualquier tarea: Sonnet 5.5 construye, Opus 5.5 revisa al momento y un enjambre de agentes Haiku 5.5 busca información e investiga, con un coste por token 80 veces menor que Fable • Fable 5.2 nunca escribe código. Solo se activa en 3 momentos: antes de cerrar un plan, cuando un error se repite y antes de dar el trabajo por terminado • Solo pagas el precio de Fable para tomar decisiones. Los modelos baratos hacen todo lo demás Así puedes montarlo: 1 → Ejecuta claude update. Fable 5.2 necesita Claude Code v2.1.321 o superior 2 → Pega el prompt de abajo en Claude Code. Primero leerá la documentación oficial de Anthropic sobre el modo advisor 3 → Configura el equipo a nivel de usuario para que funcione en todos tus proyectos: Sonnet como modelo principal, Fable como asesor, Haiku para los subagentes y Opus como revisor 4 → Hace una copia de seguridad de tu configuración y te muestra cada cambio antes de aplicarlo, comparando el antes y el después 5 → Lo que nadie te cuenta: hay una línea que le dice a Sonnet 5.5 CUÁNDO consultar a Fable. Sin ella, el asesor no se activó ni una sola vez en los 198 casos de la prueba de Anthropic El prompt completo y la documentación oficial de Anthropic sobre el modo advisor te lo dejo abajo Después, lee el artículo para exprimir Fable 5.2 al máximo:
123
573
29
5,420
684,620
Wen-Han Hsieh retweeted
Anyway, here's Wonderwall. DeformX GPU is out: an open-source CUDA solver for cables and ropes, coupled with Isaac Lab. Every string on this guitar is simulated, bending and vibrating under the robot's pick. 🔊 Sound on.
4
4
1
43
2,924
Wen-Han Hsieh retweeted
Excited to share that FINGR won Best Paper at the CoRL 2026 Dexterous Manipulation Workshop! 🏆 Huge thanks to my collaborators @QuanquanPeng03 @mattkiim6 @xiaolonw and the workshop organizers! It’s been so rewarding to bring my speedcubing hobby into robotics!
Meet FINGR: a dexterous hand that solves a 2×2 Rubik’s Cube with continuous finger tricks. Here’s how we got the fingers to work together. 🧵
9
10
4
62
6,302
Wen-Han Hsieh retweeted
Slides of my talk today: drive.google.com/file/d/1etd…
Join us this afternoon 2pm at Agentic Robotics Workshop at CoRL! Great speakers from UC Berkeley @Ken_Goldberg , CMU @GuanyaShi , NVIDIA, Gemini Robotics, Dexmate, and Tencent! And enjoy cool agentic demos! Location: JW Room 203-204 Demo: Griffin Room (2nd floor)
4
20
2
186
11,830
Wen-Han Hsieh retweeted
2
2
87
4,751
Day 1 of #CoRL2026 and the reading list is already huge! What if you can connect different works and have easy summaries. Introducing Orbis 🌌 Read, take notes, discuss with your own AI agent, and turn your library into a universe of ideas. Open source. Your research with your agents! 🧵 1/8
3
4
15
1,007
It’s so cool to see this from Anthropic! A simple action API for a strong VLM model is all u need for contact-rich manipulation. We also solved Rubik’s Cube using ASPIRE harness a month ago. Check out the code and give it a try: github.com/NVlabs/ASPIRE/tre… (and video from the e twitter post teapawt.org/darthutopian/status/21…, we did not do idle filtering at all) #CoRL2026
Replying to @claudeai
Fable 5.2 shows substantially improved skills in controlling physical robots, even on tasks it has never explicitly been trained to do. On RoboDojo, a benchmark with a simulated two-arm robot, it outperformed other models and most AI systems built to control robots.
1
1
8
9,895
Wen-Han Hsieh retweeted
Meet GR00T 2, our new generation of world-action models (WAMs): it imagines how the physical world will unfold and generates the actions that get it there. One model runs on parallel grippers, dexterous hands, and humanoids. research.nvidia.com/labs/gea…
70
205
87
1,477
125,568
Wen-Han Hsieh retweeted
Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges. They’re capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision.
3,194
9,252
6,968
96,430
21,009,650
Wen-Han Hsieh retweeted
EPO leverages differentiable energies to optimize the policy–goal pair at control rate. Unlike MoE, it enables compositional skill addition via energy-based policies without retraining the entire router. Looking forward to sharing EPO at #CoRL2026!
Planners operate on task timescales. Feasibility changes on control timescales. Bridging that gap takes a high-frequency mid-level layer. Introducing EPO: Energy-Based Policy Orchestration #CoRL2026 EPO gives each low-level policy an energy: how feasible a goal is, and how well the policy will reach it. Energies are differentiable and comparable across policies, so EPO optimizes the policy–goal pair at control rate. Unlike a learned MoE router, EPO is compositional: adding a skill means training one new energy, nothing else. It's also more precise and switches policies 7× less often, pushing the precision–robustness frontier. 📄 Paper: arxiv.org/abs/2611.02546 🌐 Project: yewr.github.io/epo 💻 Code: github.com/YeWR/EPO @wangyenjen @elvis_hsieh77 @frankzydou @ShiboZhaoSLAM @pabbeel @du_yilun @MIT_LISLab
2
6
16
557
Wen-Han Hsieh retweeted
The RL era of robotics is here, and simulation is how to scale it. An early look at our research on a achieving a better scaling paradigm for robot learning with simulation 👇
93
164
104
1,591
481,869
Wen-Han Hsieh retweeted
Planners operate on task timescales. Feasibility changes on control timescales. Bridging that gap takes a high-frequency mid-level layer. Introducing EPO: Energy-Based Policy Orchestration #CoRL2026 EPO gives each low-level policy an energy: how feasible a goal is, and how well the policy will reach it. Energies are differentiable and comparable across policies, so EPO optimizes the policy–goal pair at control rate. Unlike a learned MoE router, EPO is compositional: adding a skill means training one new energy, nothing else. It's also more precise and switches policies 7× less often, pushing the precision–robustness frontier. 📄 Paper: arxiv.org/abs/2611.02546 🌐 Project: yewr.github.io/epo 💻 Code: github.com/YeWR/EPO @wangyenjen @elvis_hsieh77 @frankzydou @ShiboZhaoSLAM @pabbeel @du_yilun @MIT_LISLab
8
19
6
100
13,289
Wen-Han Hsieh retweeted
🤖 Which robot policy checkpoint should you deploy? Lower validation loss can point you to the wrong one. Introducing 𝗥𝗔𝗖𝗘: 𝗖𝗮𝗹𝗶𝗯𝗿𝗮𝘁𝗲𝗱 𝗖𝘂𝗺𝘂𝗹𝗮𝘁𝗶𝘃𝗲 𝗦𝗰𝗼𝗿𝗶𝗻𝗴 for rollout-free checkpoint selection. RACE ranks checkpoints from a fixed set of successful reference trajectories, without running each candidate on the robot. 🧪 Simulation: stays reliable when references shift from demos to noisy rollouts, while validation loss turns negative 🦾 Real robots: 1,800 evaluations on 6 tasks. RACE is highly correlated with real deployment success rate for both Diffusion Policy and π0.5-DROID, while baselines failed. 🎬 60-second overview below 🌐 changyeon.site/race/ 🧵 1/8
3
23
57
2,042
Wen-Han Hsieh retweeted
🤖 Improving robot foundation models with real-world RL, then continually carrying that experience forward across tasks. Introducing 𝗔𝗖𝗖𝗥𝗨𝗘: 𝗰𝗼𝗻𝘁𝗶𝗻𝘂𝗮𝗹 𝗼𝗳𝗳-𝗽𝗼𝗹𝗶𝗰𝘆 𝗥𝗟 for robot foundation models. Our real-world policies learn through physical robot interaction, starting from a small set of demos. We inherit the actor–critic across tasks and distill earlier behavior, so experience accumulates while prior skills remain. 💻 𝗟𝗼𝘄 𝗮𝗱𝗱𝗶𝘁𝗶𝗼𝗻𝗮𝗹 𝗚𝗣𝗨 𝘁𝗿𝗮𝗶𝗻𝗶𝗻𝗴 𝗼𝘃𝗲𝗿𝗵𝗲𝗮𝗱: RL updates a compact actor–critic while the foundation model stays frozen. ⚡ Subsequent tasks reach 95% success in about 𝟰–𝟭𝟮 𝗺𝗶𝗻𝘂𝘁𝗲𝘀 𝗼𝗳 𝗿𝗼𝗯𝗼𝘁 𝗶𝗻𝘁𝗲𝗿𝗮𝗰𝘁𝗶𝗼𝗻. On real robots: 📉 Up to 68.5% less robot interaction vs. single-task RL 🎯 98.3% final prior-task success 🎬 A 60-second overview below. 🌐 accrue-rl.github.io/ 🧵 1/6
3
33
3
141
7,212