The greatest trick robotics startups played on VCs is convincing them it's a data problem
If your robot needs retraining for every new site, it will never profitably scale
Generalization is the only thing that matters
Did QatarEnergy brave a Red Sea shipment in an attempt to get an #LNG cargo to Poland as quick as possible, in the hope of preventing the Orlen contract termination? #natgas#TTF
How can we know if environmental review laws like NEPA deliver benefits that justify their costs?
@DanielPalken points out a natural experiment:
Some US states have their own mini-NEPAs and some don't.
So we can look at similar states that have state NEPAs (e.g., Washington) and ones that don't (e.g., Oregon), and compare them on environmental outcomes.
And... drum roll... the ones that have mini-NEPAs don't seem to have obviously better environmental outcomes.
But they do make it harder to build clean energy infrastructure.
New data shows wind and solar costs have plummeted, making them the cheapest sources of new electricity generation. Meanwhile, the claim that nuclear is the cheapest option continues to have the same problem it always has: the actual numbers. https://t.co/FQiFWWQy5L
One building model. 5,292 parts. 14 million pairs.
The trick: compare cheap boxes first, exact shapes only where boxes touch.
773 pairs left after the boxes. 9 clashes in 4 ms
We evaluated 𝗗𝗿𝗲𝗮𝗺-𝟭 via its API on RobotAgentBench, and its zero-shot capabilities are seriously impressive! 🚀
Dream-1 now 𝗿𝗮𝗻𝗸𝘀 #𝟭 overall on our VLA/WAM leaderboard. 🥇 What stood out most was its failure recovery: after dropping a fork, it tried to pick it back up. When that failed, it switched to another fork on its own.
Really curious how Dream-1 was trained to achieve this level of generalization and autonomous recovery. Would love to learn more about the recipe behind this WAM! 👀
🎉 I'm excited to present our work Shape Your Body: Value Gradients for Multi-Embodiment Robot Design as a spotlight next week at CoRL in Austin.
👉 https://t.co/HguG0mVPNq
If you wanna talk large scale RL, cross-embodiment, co-design and other fun things hit me up :D
💪 We teach humanoids where to move and how hard to apply force, using human demonstrations collected without a robot.
Introducing 🌺 FLoRe: Learning Force-Reactive Humanoid Loco-Manipulation.
Our policy learns to predict actions and forces for each hand. A hybrid compliance whole-body controller tracks these commands while the robot walks and balances.
🧵 (1/6)
So impressed by foundation LLMs skill at using RSS to tune complex robot manipulation sim parameters for geometry, dynamics, deformations, materials, lighting, texture, and camera extrinsics to match a given video.
MOSS, the litter-picking robot, is getting an SO-102 arm option!
I’ll also try @robo_nine’s modified servo firmware and run before-and-after repeatability tests to see what difference it makes.
what exactly happened here
Kostner 🇮🇹 says they upgraded to the latest gen, but this looks like a completely restyled dispenser? this is A MEGAWATT ALL in ONE?!
or did they edit the photo?! haha
#alwaysbecharging what a day :)
⚡️⚡️⚡️
The brainwashed crank @latimeralder read my post this morning so pumped out some Cfact climate denial group sponsored misinformation 😂
Is someone who isn’t blocked going to explain the primary energy fallacy to him & warn him electricity is eating fossil fuels for breakfast 👍🏻
The private sector has better incentives to adopt AI rapidly, where it is most needed, and to align it with company goals compared to the public sector.
How can #Africa’s energy priorities move from plans to financed projects? @IRENA, as Secretariat of the Global Coalition for Energy Planning & @AfDB_Group are convening 2 virtual dialogues with planners, utilities, regulators, financiers & the private sector.
✅Session 1⃣ will focus on financing IPPs, grids, decentralised energy & last-mile access.
Register now: https://t.co/liYnXfH95Q