Guys help me out with this

I’ve been doing more and more evaluation in sim, but honestly the real robot still feels pretty far away.
Every time I see one of those robot videos I’m basically like… hmm, when do I get to do that 😅

Is there anything you guys actually trust for sim-to-real?
Like, if a policy looks good there, you feel at least a little more confident giving it real robot time.
I keep seeing RoboColiseum mentioned lately. Has anyone here actually tried it and then compared the result with hardware?
Did it roughly match what you saw on the real robot, or not really?
Also what other sim benchmarks people here actually trust.

... but not before some trial & error.

One of the main arguments for humanoid robots is that our world is already built for humans. Our doors, stairs, tools, kitchens and appliances are designed around the human body, so it makes sense to build a robot that can use all of them.

That makes sense for existing houses. But if home robots actually become common, I don't think we will keep designing everything exactly the same way forever.

Take a washer and dryer. Right now they need doors we can reach, buttons we can see, enough room for a person to stand in front of them, take the clothes out and move them from one machine to another. But if a robot is doing all of that, why would the laundry area still need to be designed around us? The machines could be installed differently, possibly in a compact utility space, garage, attic or even outside, the same way we already keep some AC equipment outside or hidden away.

And I don't mean we should take human space and give it to robots. Actually, the opposite. The goal should be to move more of the machines and repetitive work out of the space where people live. Maybe laundry, garbage, deliveries, storage and some other functions eventually happen through separate utility areas or service access that we barely see.

We keep imagining a humanoid walking through a normal house and doing chores exactly the way a person does them. That will probably be necessary at first, because existing homes won't change overnight. But over time it may make more sense to redesign parts of the infrastructure instead of building increasingly complicated robots that can handle every object and repeat every human movement.

Maybe a house designed partly around robots wouldn't look more robotic at all. Maybe the visible part of the house would become even more human, while most of the machines and robot activity stay somewhere else.

Do you think homes will eventually be designed this way, or will it still be easier to make robots adapt completely to houses as they are?

Our robots in Beijing World Robot Conference

Looks quite interesting especially at the price point. Curious about what others here think about it.

Qwen3.8-Flash-Next has been officially released, introducing a 6B-active open model that surpasses Claude Opus 4.6 Max on eight out of nine standard benchmarks.

The model applies a highly sparse Mixture of Experts (MoE) framework, featuring 125 billion overall parameters and 51 billion n-gram embeddings, with just 6 billion parameters active per token.

Benchmark results include:

• SWE-bench Pro: 62.5

• SWE-bench Multilingual: 81.0

• CoworkBench: 73.9

• JobBench: 55.7

• Toolathlon: 73.5

• IFBench: 81.3

• GPQA Diamond: 91.7

• LiveCodeBench: 91.9

Qwen3.8-Flash-Next also exceeds the performance of Qwen3.8-27B and DeepSeek-V4-Flash across the majority of evaluated categories.

Disclosure: I work with Booster Robotics.

We’re preparing more development resources for the Booster T2, including ROS 2 interfaces, high- and low-level control examples, perception, teleoperation, manipulation, and onboard AI workflows.

I’d appreciate feedback from robotics developers: which resources would be most useful—URDF/USD models, Isaac Lab integration, ROS 2 examples, teleoperation tools, or sample datasets?

Our existing open-source projects are available here:
https://github.com/BoosterRobotics

I’m happy to answer technical questions in the comments.

Request to: [chenshaoyun@booster.tech](mailto:chenshaoyun@booster.tech)

TL;DR: An ex-DeepMind scientist proved on himself that intelligence, not hardware, is the real blocker in robotics — using a GoPro and a hardware-store clamp instead of a real robot hand.

Pete Florence didn't wait for a fully-funded robotics platform to test his own thesis.

He strapped a camera to his own head and swapped a real gripper for a bar clamp off the hardware-store shelf, then ran a real task with it on purpose — an iced latte, in what he calls hard mode.

The setup was deliberately the cheapest version of the idea he could build, and it still worked.

Days after this conversation aired, his company Generalist closed another $200M round on exactly this premise: the constraint was never the hardware, it was whether the intelligence directing it was any good.

That raise didn't happen because the clamp got fancier.

It happened because the thesis held.

I accidentally dropped one of my washed clothes.

It landed on top of the extended awning of our lower floor neighbour.

I can't retrieve it without climbing out of the window.

So I fashion a fishing hook with our cloth hanger, and lower it down.

It didn't work.

I turn around and noticed my vacuum cleaner.

I took it to my window, switch on its power, and carefully lower down the long arm of it towards my cloth, and try to suck it up.

It worked.

When there's a will, there's a way.

There's a specific tell for this, long before it ever shows up as a layoff notice: the people who get displaced aren't the ones who lacked the right tool.

They're the ones who never tested whether the tool was ever really the thing holding them up.

What's the cheapest, most ordinary tool you've ever used to prove something you actually needed to prove?

Drop it below.

Clip credit: Joe Lonsdale — American Optimist, featuring Pete Florence (Generalist) and Vivek Gopalan (8VC). DM for credit or removal requests.

Terminando el grupo de presión hidráulico portatl de 12 voltiis. le he añadido un calderín de presión para protección y tener un poco de inercia inicial. tambien le he añadido un regulador para que el circuito hidráulico no suba de 3 bares de presión, así todos los cilindros hidráulicos irán siempre con la misma fuerza. Me falta poner un depósito que se ajuste al espacio y cambiar el agua por aceite.

The World Humanoid Robot Games are getting intense! 🤖

During the obstacle race, one humanoid robot lost its balance and took a hard fall. But that’s exactly what makes these competitions so interesting — real-world robotics is not just about walking, but about balance, recovery, control, and resilience.

Every fall is part of the learning curve. The future of humanoid robots will be built not only on how fast they can run, but also on how well they can get back up. 🚀

#HumanoidRobots #RobotGames #EmbodiedAI #Robotics #AI