
Research @NVIDIA | cuRobo Lead | views my own
San Francisco, CA
Joined October 2018
- Tweets66
- Following742
- Followers453
- Likes170
Pinned Tweet
In 2022, our first cuRobo prototype took 100 ms and consumed the whole GPU.
Today, we’re open-sourcing cuRoboV2-Fast under Apache 2.0: collision-free, torque-limited motion planning in 4.4 ms.
Collaborating with researchers across robotics and GPU architecture brought us to 50 ms, then 30 ms. B-splines, GPU time optimization, and faster kernels brought us to 4.4 ms while producing motions that execute faster.
As planning got faster, perception, grasping, and control became the bottlenecks. We built a 7 ms perception pipeline (SceneFields): from stereo RGB to depth, voxels, and robot and object segmentation, and accelerated GraspGenX to 1.6 ms. We also integrated wrote a cpp controller for YAM and built a low-latency CUDA-GL visualizer for an interactive live view of the task.
We deploy the stack as a two-process application: one dedicated to real-time control, the other handling perception, grasping, planning, and visualization.
On a real YAM arm, pick-and-place cycles averaged 6 seconds, 34.6% shorter than cuRoboV2.
A long journey, made possible by amazing collaborators. The rest of the stack is coming soon.
research.nvidia.com/labs/lpr…
Joint work with @Adithya_Murali_ , @BirchfieldStan
Prompt your agent with :
Use cuRobo cpp_runtime for collision-free motion planning: github.com/NVlabs/curobo/blo…
Our planner can clear novel objects from a table, avoids collisions between the grasped object and the scene during transfers. [Videos playing at 1x]
1
4
502
Torque sensing is coming to every robot...
Introducing Silica.
We make actuators 50x more sensitive at an affordable price, so that robots can feel the world around them.
When I was at Apple, I developed technology that helped humans live more creative and connected lives.
Now I make hardware for the training age, so that robots can begin to share that world with us.
We just opened pre-orders. Check out silicamachines.com to learn more.
4
159
cuda.core greatly simplifies integrating kernels with pytorch. We did this for curobo and installation went from 10 minutes to a few seconds. See here for an example: github.com/NVlabs/curobo/blo…
Major ecosystem libraries like PyTorch and CuPy now build on the shared cuda.core foundation, reducing version conflicts and interop bugs for developers.
With CUDA 13.3, @nvidia released CUDA Python 1.0, the libraries and tools that give you the full CUDA platform from Python. Using standard CUDA context, devices, streams, and memory are shared with the rest of the Python GPU ecosystem, so your own kernels can run against CuPy arrays or PyTorch tensors without copying data.
Read the full blog post: bit.ly/4hcSu2C
11
505
Balakumar Sundaralingam retweeted
🎉 CUDA Rust is here! Team 🟩 ❤️ 🦀
You can now write CUDA kernels in plain Rust, two ways: cuda-oxide (SIMT) and cutile-rs (Tile).
developer.nvidia.com/blog/in…
Balakumar Sundaralingam retweeted
We are giving out $500k and free YAM arms to academic teams to design dexterous manipulation benchmarks and run real-world evaluations.
General-purpose robots could soon be in homes and data centers. Society should know what robots can do before they get there.
Congrats! Truly autonomous robot that cleans messy homes!
Replying to @maticrobots
@maticrobots shipped its 10,000th robot this week. 🤖
Back in 2018, there was no product - just a small team, a lot of ideas, and an unreasonable amount of optimism.
The photos below are from those early days: prototypes, experiments, customer interviews, design reviews, and plenty of things that never made it into the final product. We first started shipping in November 2024.
Huge credit to the Engineering and Production teams that made the 10K bots happen!
Great engineering gets you to the first unit.
Great teams get you to the first 10,000!
Still a lot more to be done. 🚀
1
177
Open Source Grasp Generation for any gripper, any object. You can pair it with cuRoboV2 & Newton for large scale grasp trajectory generation. Try it out here: github.com/NVlabs/GraspGenX/…
Replying to @nvidia
@NVIDIA is working on one of the hardest problems in Physical AI so you don’t have to: generalist robotic pick-and-place.
We are excited to introduce GraspGenX at #CVPR2026—a foundation model for robotic grasping that works out of the box for unknown robots, novel objects, and unseen environments.
Unlike Vision-Language-Action (VLA) models or dedicated grasp networks that require expensive, embodiment-specific training, GraspGenX is cross-embodiment and works zero-shot. You simply pass a "robot prompt" alongside an image of the object to generate actions.
🚀 Key Highlights:
1) Scaling: Trained on over 2 Billion 6-DoF grasp rollouts entirely in physics simulation—a dataset size practically impossible to collect via real-world teleoperation.
2) Zero-Shot Transfer: Works out of the box for several common robot grippers widely used across the research community and industry.
3) Built for the Agentic Era: Features native MCP support, client-server architecture, and skills.md, allowing seamless integration into LLM/Agentic robotics workflows.
4) Full Pipeline Integration: Pair it with other open foundation models (like SAM3) and advanced motion solvers like cuRoboV2 for full deployment in entirely unknown environments.
If you are currently executing pick-and-place with a VLA or WAM, you can use GraspGenX to generate sim-verified trajectory data and inject it into your pipeline. No need to waste precious real-world engineering hours on data collection for standard manipulation tasks.
🌐Website: graspgenx.github.io/
💻Code: github.com/NVlabs/GraspGenX
📄Paper: arxiv.org/abs/2606.00998
📍CVPR Booth: Poster 619 on Jun 6 1:45 session at ExHall F
This work was led by the incredible @BeiningH (Princeton), in collaboration with a phenomenal team at NVIDIA: @erwincoumans, @yu_wei_chao, @balakumar_, @clembow, and Stan Birchfield
#CVPR2026
27
3,249
Balakumar Sundaralingam retweeted
Let's meet and chat at #CVPR!
Come by our poster or schedule a coffee chat🙌
Fri afternoon:
- Fast-FoundationStereo: nvlabs.github.io/Fast-Founda…
- CARI4D: nvlabs.github.io/CARI4D/ (w/ @XianghuiXie )
Sat morning:
- Event6D: chohoonhee.github.io/Event6D… (w/ Jae-Young)
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
Balakumar Sundaralingam retweeted
Use both:
web.mit.edu/dimitrib/www/Les…
TO is a Newton step on the Bellman equation. Policies and value functions are "memories" of past solutions; TO should be optimizing over them at inference time. Best of both worlds. Some of the strongest RL methods do this.
Should legged robot use RL or trajectory optimization? #icra2024
Balakumar Sundaralingam retweeted
New work on vectorizing geometric fabric controllers for RL workflows at scale. DeXtreme: Fabric Guided Policies (FGP).
Policies are hard on hardware. We need low-level controllers at deployment, which means we need them during training.
FGPs increase hardware lifetime, enable quick iteration on training and deploying policies, and allow us to inject useful inductive bias into the system.
Our code for CUDA accelerated motion generation is out! Supercharge your workflows with fast batched robotics modules, including kinematics, collision queries, optimization, and motion planning. #PyTorch #Nvidia #Robots
🤖 cuRobo, a new #CUDA accelerated motion generation toolkit, can solve complex #robotics problems in milliseconds. ⚡
It includes implementations of kinematics, collision checking, numerical and trajectory optimization, and more.
👀 #NVIDIAResearch code nvda.ws/3MxDmNG
Balakumar Sundaralingam retweeted
DeXtreme is our new work on scaling sim-to-real for contact-rich manipulation with a vision-based state estimation on a robot hand with the infrastructure we have been developing with Isaac Gym over the past one year.
arxiv.org/abs/2210.13702
dextreme.org/
Balakumar Sundaralingam retweeted
I'm excited to share our #RSS2022 workshop paper: Neural Motion Fields: Encoding Grasp Trajectories as Implicit Value Functions.
w/ @Adithya_Murali_ , @balakumar_ , Wei Yang, @animesh_garg, and Dieter Fox at @nvidia.
Paper: arxiv.org/abs/2206.14854
Video: youtu.be/B-pEhT1pi-Q
Balakumar Sundaralingam retweeted
Factory: Fast Contact for Robotic Assembly, our recent work, is a set of simulation methods & robot learning tools for contact-rich interactions for robotic assembly. It will be presented at RSS next month.
Paper: arxiv.org/abs/2205.03532
Website: sites.google.com/nvidia.com/…






















