Go @OlympusMonsTeam! Go @PALRobotics!💪🏽💪🏽 2 years of hardwork turned to a dream come true!! Very proud to be part of the team!

Mowing some lawns for the next RoboPlan release... (among many other things, including performance improvements)

same robot but they gave me a gpu 🏃

#ICRA2026 parade of robots 🤖

The latest haptic glove from LeRobot's @nepyope looks alive 👀 Homunculus F³ drops next month. Repost if you're hyped for new #lerobot hardware 🤗 #HF #lerobot #haptics

The agility determination to have their digits on the shop floor is paving the way for a more flexible and responsive supply chain🤖🦾

I think more people should be teaching their robots how to fall. incredibly important

cuRobo is now open source under Apache 2.0. cuRoboV2 adds <1 ms GPU-native TSDF/ESDF semantic mapping, whole-body IK/MPC and trajectory optimization for humanoids, ~50 ms torque-limited planning. Report: arxiv.org/abs/2603.05493 Code: nvlabs.github.io/curobo

SMP is such a brilliant solution to AMP's issues.🏆 AMP: Binary signal (real/fake). Fantastic for one to few styles, but you're always fighting the discriminator learning too fast resulting in zero gradient. It's also super intolerant of style imbalance in dataset. SMP: Continuous signal (fancy diffusion noise trick to measure distance against learned plausible motions). Every single episode always provides valuable reward signal. No signal collapse. No co-evolution balance pain. More forgiving of imbalanced motion types in dataset. Wild. yxmu.foo/smp-page/

Two Anthropic engineers, who built Claude just explained why you use less than 10% of actual Claude abilities. This 24-minute talk will change how you use Claude Code forever. Watch it, then read the breakdown below👇

This is how you navigate the Asimov v1 repo.

I trained a grasping policy on objects of various shapes and sizes using the new mjlab feature and this really cool pivot grasp strategy emerges for large flat objects that are larger than the gripper aperture. So cool and beautiful to see!

This is the most exciting #PlotJuggler release in a long time: 3.17.0! There are so many things I am exited about, but the most important one, is the integration with Mosaico (link below). Now PJ can download data directly from the cloud. Thread 🧵👇

We're open-sourcing Asimov v1's mechanical design on Monday 🧡

Claude can actually do CAD now in @Onshape Here it worked for an hour and built a 4-part monitor arm, starting only from a sketch and description. The trick was to give it the tools to look at its own work. Introducing: Jarvis Onshape MCP

FlashSAC is out — when XQC meets SimbaV2 massively parallel simulation. Stable off-policy RL for high-dimensional robot control. 50 tasks, 10 simulators, sim-to-real humanoid walking in just minutes. Very proud to be a part. CrossQ-style batch norm keeps winning in deep RL 🚀

Posted v1.4 roadmap for mjlab: linkage modeling, finding and crushing all NaNs, per-world meshes, and better RL-library support. If something on the list should move up, or I've missed something important, comment on the issue: github.com/mujocolab/mjlab/i…

FlashSAC is a new RL method for robotics that fixes a core problem PPO is stable but wastes data Off-policy methods reuse data but are slow and unstable FlashSAC combines both advantages fast training stable learning even in high dimensional robots

Headline new features from MuJoCo 3.7: DC motor actuator, midpoint integrator for super stable free bodies, and improved equality constraints. mujoco.readthedocs.io/en/sta…

We’re hiring a Reinforcement Learning Engineer 🤖 Join us to tackle real-world challenges with humanoid robots. Apply or reach out! linkedin.com/jobs/view/44015…

$5,900 — Unitree just put its next humanoid into global sale🤖 After G1 quietly became one of the most in-demand humanoid platforms, Unitree is now pushing R1 overseas via AliExpress, targeting North America, Europe, Japan, and Singapore. Same company, different playbook: lower price, faster scale. R1 comes in at $5,900, with a lighter AIR version at $4,900 — a sharp drop from G1’s 99,000 RMB (~$16K ) entry point. It’s a 123 cm humanoid, weighing about 29 kg, with up to 26 DOF across legs and arms. The lighter R1 AIR drops to 27 kg with 20 DOF, built for more agile, lightweight deployment. This isn’t a concept rollout. Unitree already went through the demand phase last year. Globally, G1 has already secured strong demand across research, education, and commercial applications, with supply struggling to keep up at one point. Now they’re catching up. Unitree says its humanoid shipments crossed 5,500 units in 2025, with total production above 6,500 units. Orders were higher than deliveries — meaning backlog, not lack of interest. That context changes how R1 lands. This isn’t about “another robot launch.” It’s the next unit in a pipeline that’s already selling, scaling, and shipping. G1 proved people want humanoids. R1 is what happens when you try to meet that demand — cheaper, broader, global.