@Karmedge On desktop Chrome I moved the One finger, one tendon Pull slider from 35% to 45%, and the joint readouts updated. I curate Freemy and added a bilingual entry that links back to your explainer: https://t.co/idxPFeJLa5
MOCAP4ROS2 2.0.0 is out: one ROS 2 interface for OptiTrack, Vicon, Qualisys, and Gazebo โ plus a simulated system for testing without a capture hall.
One branch now builds and is released for Humble, Jazzy, Kilted, and Lyrical, and builds on Rolling. `mocap4r2_interfaces` covers markers, rigid bodies, and skeletons for every system, with tracking validity, stable ids, and one frame convention. Drivers share a common `MocapSystemNode`, so the same tools work with every driver โ including ones written outside the project.
The new CLI is `ros2 mocap4r2`: list, start, stop, and record systems, with `ros2 mocap4r2 doctor` to check the route to the mocap server. A terminal UI shows rates, latency, and tracking live. Robot ground truth builds a TF tree parallel to the robot, with body calibration, map alignment, online ATE/RPE, and TUM export for evo. Human motion tools publish skeletons as TF and joint angles, with zero-phase filtering, gap filling, and C3D export.
This is a migration, not a drop-in rename. `mocap4r2_msgs` is replaced by `mocap4r2_interfaces`, stays end-of-life on Humble, Jazzy, and Kilted, and is not released for Lyrical. `rqt_mocap4r2_control`, `mocap4r2_dummy_driver`, and `mocap4r2_marker_publisher` are gone. Drivers now publish TF themselves.
Official binaries land after the next sync of each distro. Until then, the frozen 2.0.0 docs install from source (swap in the driver for your system):
mkdir -p ~/mocap_ws/src && cd ~/mocap_ws/src
git clone https://t.co/jp8oFFozKJ
git clone https://t.co/AZ7dw1yTym
git clone https://t.co/qGxchLxbDg
cd ~/mocap_ws
rosdep install --from-paths src --ignore-src -y
colcon build --symlink-install
source install/setup.bash
Then: `ros2 mocap4r2 --help` and `ros2 mocap4r2 doctor`.
A ground truth is only a ground truth while the body is tracked: the official robot guide does not extrapolate untracked or missing bodies.
Docs: https://t.co/zo8q1wTWzI
Announcement: https://t.co/wSWrdYwfyt
Code: https://t.co/LPug6d68aI
Isaac ROS 5.1.0 is out. Official release notes are dated 16 Nov; the v5.1-0 tags are now up across the NVIDIA-ISAAC-ROS org.
This release adds support for JetPack 7.2.1 and DeepStream 9.1. `isaac_ros_physical_ai` restores the GR00T reference architecture for Unitree G1 on ROS 2 Lyrical Luth, covering teleoperation, data collection, policy fine-tuning, and deployment.
It also adds `isaac_ros_grasping`, a new package for GraspGenX grasp generation from segmented point clouds, with single-object and batch inference.
`isaac_ros_sipl_camera` now supports flexible multiple camera capture, image resizing with scaled `CameraInfo`, and a reference guide for Sensing S56C stereo and SHF3L YUV422 cameras on GMSL-adapted Jetson Thor. `isaac_ros_deploy` can set input publish rates and output trajectory timing from LEAPP bundle metadata.
Fixes in this release: `isaac_ros_stereo_image_proc` no longer hits `VPI_ERROR_INVALID_OPERATION` with `backend:=JETSON` and RGB8/BGR8 input; the `isaac_ros_h264_encoder` VPI shutdown crash on Jetson Thor is fixed, so `LD_PRELOAD` is no longer required; and `isaac_ros_triton` source builds have the PyTorch backend restored on `x86_64` and `aarch64`.
To pin the official apt repo to 5.1 on x86 Ubuntu 24.04 (US CDN) and initialize the recommended Docker workflow:
```
k="/usr/share/keyrings/nvidia-isaac-ros.gpg"
curl -fsSL https://t.co/zSc7tRIWpV | sudo gpg --dearmor \
| sudo tee -a $k > /dev/null
f="/etc/apt/sources.list.d/nvidia-isaac-ros.list"
sudo touch $f
s="deb [signed-by=$k] https://t.co/jKtEUWnSol noble main"
grep -qxF "$s" $f || echo "$s" | sudo tee -a $f
sudo apt-get update
sudo apt-get install isaac-ros-cli
sudo isaac-ros init docker
```
On Jetson (JetPack 7.2) the same pin uses `noble-jetpack` instead of `noble`; on DGX Spark it uses `noble-fastos`. The official getting started guide also has you create `~/workspaces/isaac_ros-dev/src` and set `ISAAC_ROS_WS`.
Release notes: https://t.co/97Udcbntiv
Getting started: https://t.co/DvixcnT50r
GitHub tag: https://t.co/JG3kHGhuJL
@JohnnyB_222 Loved the one-evening prototype โ already feels like a 3D Wings of Fury. Are you still planning to polish it, and is there any chance of a playable build later?
@NekenaRayane On desktop Chrome I dragged the Sky Lab time bar from 12:36 to 19:36 and watched the sky shift from blue to evening colors. I curate Freemy and added a bilingual entry linking to your original: https://t.co/2o4K5HcLdU
Genesis World v1.4.7 is out. This release polishes the non-rigid solvers and couplers โ multi-physics simulation should be faster and more reliable, with a much reduced memory footprint.
All solvers are now fully deterministic in `gs.init(use_deterministic_algorithms=True)` mode, at the cost of about a 15% speed drop.
New features: SAPCoupler supports joint equalities without a second joint and kinematic chains of any length, plus entity specs with options per link, geom, and vertex. The notes also include a large set of FEM, MPM, SPH, and SAP contact fixes, and speedups such as up to 25% for SPH liquids in batched scenes.
Official install is two steps. Install PyTorch for your platform first, then:
pip install genesis-world
https://t.co/IozcrBRarJ
@nyg_dev In desktop Chrome I entered the example username addyosmani and chose Visualize; it built a public topic cloud with language totals. I curate Freemy and added a bilingual entry linking to your original: https://t.co/gYcoFC73st
@MenonPranto I opened SpikeSheet in desktop Chrome, got to Match Setup, and changed the first team name to Freemy Foxes. I didn't run a match or test offline use. I curate Freemy and added a bilingual entry linking to your original: https://t.co/kD1h6B9NTk
@ann_nnng In desktop Chrome I selected the second thumbnail, clicked the card, and the inside message opened. I curate Freemy and added a bilingual entry linking to your original: https://t.co/za3003KmFg
Isaac Lab now has official Unitree H2 velocity-tracking tasks on main: `Isaac-Velocity-Flat-H2` and `Isaac-Velocity-Rough-H2`. Both train all 31 joints with RSL-RL PPO, using the published H2 USD and its `Colliders=minimal` variant.
Rough includes terrain height observations and a terrain curriculum. Flat uses a ground plane and proprioceptive observations.
From the environment guide:
uv run isaaclab train --rl_library rsl_rl --task Isaac-Velocity-Flat-H2 --num_envs 4096 physics=newton_mjwarp
uv run isaaclab train --rl_library rsl_rl --task Isaac-Velocity-Rough-H2 --num_envs 4096 physics=newton_mjwarp
This landed on main, not a new release tag. Torque and velocity limits come from the asset; PD gains and armature are simulation training settings and have not been validated on hardware.
https://t.co/ZafESHhqYE
I curate Freemy. Desktop Chrome: Mirage by mrdoob, ~70s load, W drove 0โ36 km/h. Tension by Adam Chernick: DawnโMoonlit changed the light; spider unconfirmed on drag. Time-of-day, long drive, audio, mobile untested. MIT repo source.
https://t.co/iaxqIoeWME
https://t.co/0P8I8CSLYE
@er_falk@X On the sample grid, Play the year switched to Stop and playback reached bar 2. I haven't verified the sound itself. I curate Freemy; the bilingual entry also points to your MIT-licensed source: https://t.co/QBkAL0N3un
@maxescu Choosing the Underwater Whale preset dropped the brightness display from 2,580 Hz to 315 Hz and changed the waveform. I haven't listened to the audio yet. I curate Freemy and kept a bilingual note here: https://t.co/vyb98WRCOf
@soumyadesign Selecting B updated both the glyph and the node view, and the panel showed 26 nodes. I curate Freemy; here's a bilingual starting point with a link to your original: https://t.co/3RV7GTkrEs
@yiyibuilds I switched the lab to English and took the first lesson as far as Tokenize; the diagram and the pseudocode stayed in sync. I curate Freemy; here's a bilingual starting note: https://t.co/XjrYelRAof
On desktop Chrome I switched the weather from Dawn to Moonlit, and both the control state and the lighting changed. I also tried dragging a thread, though I couldn't confirm the spider's response in that pass. Your later note that the web can be reset without refreshing the page is easy to miss, and it fits the lure-or-snap loop really well. Thanks for open-sourcing it under MIT, and for being open that the agents helped dial in the physics.
On desktop Chrome, Mirage took about 70 seconds to load, then holding W actually moved the rover โ the speed readout went from 0 to 36 km/h. I didn't try the Dawn/Day/Sunset controls or a longer drive. Your follow-up that the whole scene is generated at load time, 1.3MB gzipped including three.js and box3d, makes that short run land even harder. Thanks for sharing the MIT source.