Sharpa is an AI robotics company dedicated to developing ultra-high performance robots and core components. We Manufacture Time by Making Robots Useful.
United States
Joined April 2025
- Tweets371
- Following199
- Followers6.2K
- Likes1.2K
Sharpa retweeted
Can we make dexterous robots compliant without sacrificing precise control?
Excited to introduce FOMO-Dex: Force-Modulated Dexterity! 🖐️
FOMO-Dex brings force-aware compliance to dexterous teleoperation, enabling robots to pick up thin cards and chips 🃏, cut paper 🔪, turn nuts 🔩, and use screwdrivers 🪛.
These force-aware demonstrations also train better autonomous policies!
🎥 Videos & paper: yingyuan0414.github.io/fomo-…
🧵👇 (1/8)
Sharpa retweeted
Its a good mission statement! And an impressive robot
Episode 1 — We manufacture time by making robots useful
This August, North steps behind the counter at a real DQ store in Shanghai; and it will be a full shift, with real customers, real orders, and of course, real ice cream.
@GoingBallistic5 nailed it: North isn't a replacement for workers, it's a "substitute," filling in or augmenting the workforce where it's needed.
The North robot adds capacity when teams are understaffed, late shifts are hard to fill, or repetitive work keeps people from more valuable tasks.
@Hmorvaridi caught exactly what we meant with the football scene: people enjoy the match while North takes the shift.
@roydendsouza linked it back to our mission: We manufacture time by making robots useful. Thanks, Royden, for pointing out that it is "the best worded mission" you've seen in a long time.
Stay tune for the next episode!
#Sharpa #SharpaNorth #UsefulRobots #EmbodiedAI #AIRobotics #overthehorizon
Sharpa retweeted
Can't wait for all the 'cool' stuff to begin at Dairy Queen in Shanghai! 🍦❄️⛄️
@SharpaRobotics
@GoingBallistic5 @Hmorvaridi @anatomyumea 🔥🦾
Episode 1 — We manufacture time by making robots useful
This August, North steps behind the counter at a real DQ store in Shanghai; and it will be a full shift, with real customers, real orders, and of course, real ice cream.
@GoingBallistic5 nailed it: North isn't a replacement for workers, it's a "substitute," filling in or augmenting the workforce where it's needed.
The North robot adds capacity when teams are understaffed, late shifts are hard to fill, or repetitive work keeps people from more valuable tasks.
@Hmorvaridi caught exactly what we meant with the football scene: people enjoy the match while North takes the shift.
@roydendsouza linked it back to our mission: We manufacture time by making robots useful. Thanks, Royden, for pointing out that it is "the best worded mission" you've seen in a long time.
Stay tune for the next episode!
#Sharpa #SharpaNorth #UsefulRobots #EmbodiedAI #AIRobotics #overthehorizon
Sharpa retweeted
Okay this is actually wild. Google gave two robots the same task. Nobody told them who does what. They split the work, coordinated, and cleaned the room together.
Gemini Robotics 2. Published today.
A humanoid walks to a table, picks up a watering can, steps to a shelf, crouches, and places it in the right bin. One spoken instruction.
Another ties a trash bag knot with five fingers. Seals a ziplock bag. Screws in a lightbulb. 22 degrees of freedom per hand. Figured out the hand movements itself.
→ Adapts to a new robot body in hours with less than 200 examples
→ Runs locally.
→ Same model controls humanoids, dual-arms, and grippers
→ Self-corrects when a step fails
For 50 years the hardest problem in robotics wasn't walking or seeing. It was hands. The fine motor control every two-year-old does without thinking. Google just published a model that does it.
And taught robots to work as a team while they're at it.
Sharpa retweeted
The world is designed for humans ♥️. #GeminiRobotics 2 gives humanoids the intuitive and intelligent whole body coordination we take for granted so they can be. more helpful!🦿 🧠
@GoogleDeepMind #Robotics
Sharpa retweeted
Gemini Robotics ER 2 is our most capable embodied reasoning model designed for physical AI 🤖
Built as a high-level brain for robotics, the model connects directly to the Gemini Live API. It processes continuous video streams to track progress, call tools, search the web, and command an action model in real time.
Upgrades for developers include:
+ Enhanced progress tracking
+ Real-time mid-execution failure detection
+ Multi-robot collaboration
+ Advanced safety instruction following
Sharpa retweeted
Really cool: Google DeepMind says Gemini Robotics 2 can control a humanoid’s entire body, from walking and crouching to manipulating objects with five-fingered hands.
The system combines three models:
Gemini Robotics 2 converts vision and instructions into movement.
-ER 2 plans tasks lasting several minutes, tracks hundreds of decisions and coordinates multiple robots.
-On-Device 2 runs locally and can adapt to a completely new two-arm robot with fewer than 200 examples collected over a few hours.
The same Robotics 2 checkpoint controlled different versions of Apptronik’s Apollo humanoid and a Franka Duo robot.
However, performance remains uneven: Success rates ranged from 45.7% to 76.3% on whole-body tasks. Five-finger manipulation ranged from 32% for using a dustpan to 92% for unscrewing a bulb.
Sharpa retweeted
Gemini Robotics 2 is the first action model to combine *general* whole body control and multi fingered dexterity, controlling more degrees of freedom than any other action model before - full body humanoid for stepping / squatting and high DOF hands for dexterity - all with one checkpoint and one set of weights, that which is also SOTA for controlling bimanual grippers.
Sharpa retweeted
Gemini Robotics 2 is out today and is a step function change in robotics! It is the most advanced and general humanoid intelligence out there.
Some key highlights of the model 🧵:
Sharpa retweeted
Meet Gemini Robotics 2: our next-generation physical AI from @GoogleDeepMind that gives robots the ability to think, reason, act, and even work together.
Sharpa retweeted
For decades, we’ve dreamed of robots that can seamlessly step into our world and lend a hand.
Today, we take a major stride toward making that dream a reality:
Introducing Gemini Robotics 2 from @GoogleDeepMind, the intelligence layer powering the next generation of truly adaptable robots. This major advance unlocks intelligent whole-body control, advanced dexterity, and even multi-robot collaboration 🤯.
Ok but... how does a robot actually "think"?
Real-world tasks take time and planning. To manage that complexity, our new embodied reasoning model, Gemini Robotics ER 2, acts as the robot’s high-level brain, enhancing the robot’s capabilities to:
— Observe the environment
— Reason about the actions needed to complete the task
— Coordinate with the vision-language-action model to carry out actions
— Track progress until the job is done
This setup allows robots to execute complex multi-step workflows, self-correct if a step fails, and adapt to completely novel situations.
Learn more about Gemini Robotics ER 2 (and our two other brand new models) here: goo.gle/4x4E8q6
Sharpa retweeted
Introducing Gemini Robotics 2: the next era of truly adaptable robots from @GoogleDeepMind 🤖
Powered by three new models, it brings whole-body control, fine dexterity, and teamwork to complex tasks — making robots more helpful in the real world. xitter.cf/GoogleDeepMind/s…
Sharpa retweeted
Whole-body reasoning for humanoids! 🦿
@GoogleDeepMind released Gemini Robotics 2, three models working together to handle what actual robots need.
Gemini Robotics 2 (the VLA) converts vision and language into action.
For the first time at scale, it controls the entire humanoid body as one system, not arms separately from legs.
The robot reasons through whole-body movement while walking, crouching, bending.
Advanced dexterity on 22-degree-of-freedom hands.
Multi-robot collaboration. A humanoid and a bi-arm robot can work together on the same task. Different embodiments, shared understanding. One robot can't reach a space, the other can. They communicate and divide the work.
Then, the Gemini Robotics ER 2 is the embodied reasoning model, the robot's high-level brain. Plans multi-step tasks. Talks to humans. Handles uncertainty. Can request human intervention when it's not sure.
Last but not least, safety!
ASIMOV2 benchmark shows ER 2 is the safest robotics model to date.
New ASIMOV-Agentic benchmark tests agentic behavior under uncertainty, can the robot recognize when to ask for help instead of guessing?
That's soo cool! 😎
Congrats @parada_car88104 and team! 🦾
Read more here: deepmind.google/blog/gemini-…
~~
♻️ Join the weekly robotics newsletter, and never miss any news → ziegler.substack.com
Sharpa retweeted
Perhaps the most impressive example of robot dexterity out there, and you can download the data now ->
The task of the #IROS2026OrigamiChallenge is to fold a traditional paper airplane in 6 folds. It's one of the hardest manipulation benchmarks around: paper is thin, deformable, easy to occlude, and unforgiving of contact timing.
A successful policy has to coordinate two arms, two dexterous hands, tactile feedback, and multi-view vision over a long sequential horizon.
Sharpa retweeted
@GoingBallistic5, @roydendsouza, and @Hmorvaridi putting this much faith in our design and operation? No pressure 😅 Bar's set, now we deliver. Shanghai, August: see you there 💪
⚡SHARPA LEAVES THE LAB, HEADS TO DAIRY QUEEN
@SharpaRobotics is taking its 22-DoF Wave hand beyond demos with Project Edison, preparing its North robot for a real Dairy Queen deployment in Shanghai.
@GoingBallistic5 got an early look at the project.
A few things that really stood out from our conversation:
> Sharpa is building the full stack around the hand, including the robot arms, mobile base and AI
> North has to work in the existing kitchen, with no redesign of the workspace around the robot
> Scott Walter says making a Blizzard involves roughly 50–60 steps through an existing human workflow
> The preview footage is teleoperated and sped up. Sharpa expects the released system to be fully autonomous
But the bigger story here is Sharpa taking dexterous robot hands from impressive demos to complete systems capable of useful commercial work.
And Sharpa founder David Li has offered to join us on Over The Horizon to take us deeper inside Project Edison.
Great discussion with @GoingBallistic5 and @Hmorvaridi.
Looking forward to hopefully having @SharpaRobotics and David Li join us for a deeper dive into their work!
⚡Full episode linked below in the thread!👇🏽
Sharpa retweeted
Dairy Queen just hired its first robot worker
The 86-year-old soft-serve pioneer is partnering with the Singaporean startup Sharpa Robotics to automate ice cream service. Sharpa's humanoid robot, North, is training to whip up Blizzards autonomously at DQ's new store in Shanghai starting in August 2026.
The AI-powered robot will use its highly dexterous five-fingered SharpaWave hands for the full end-to-end process. Each hand has a reported 22 degrees of freedom (DoF), which approaches the dexterity and range of motion of human hands. They have more than 1,000 tactile pixels per fingertip with very high sensitivity for real-time force feedback, texture detection, and precise pressure control.
Launched in 2024, Sharpa says it's already started mass producing its robotic hands that have gained international attention.
The hands are priced starting around $52,000 apiece. Sharpa has not disclosed pricing for its full-sized humanoid.
Sharpa retweeted
Stay tuned, folks!
Coming Soon: Full breakdown of the Sharpa hand with @GoingBallistic5
👇🏽🔥🦾🖖🏽
Special Scoop! 🍨
Sharpa granted exclusive access to Project Edison, its new partnership with Dairy Queen.
Sharpa is more than just hands. It's now a full-stack AI robotics company — and today it rolled out North, equipped with its state-of-the-art hands, into a consumer-facing commercial setting for the first time.
Working within human infrastructure
The point isn't to replace workers — it's to substitute as needed. That constraint has teeth: Sharpa can't just reshape the workscape to make it easier. The robot must operate within the store's existing layout, tools, utensils and appliances.
Take the mixer. It's harder than it looks. The ice cream is lumpy and viscous, producing unpredictable torques and forces the arm and hand must counter in real time — all while translating the cup vertically without crushing it.
No special cup dispenser, either: North uses the same one humans do, with the same challenge of pulling exactly one cup at a time. Many of these steps would be difficult, if not impossible, with pinch grippers.
Long-horizon tasks
Long-horizon tasks remain the bane of many autonomous systems. While this demo was teleoperated, Sharpa expects these tasks to run fully autonomously by launch.
What's missing
The video is sped up only 1.5x — still short of average human speed. Sharpa expects cycle times to improve over the coming months, but they're quite not there yet.
The road to launch
The target: full autonomy by the official launch in August at the Dairy Queen in Shanghai. One flavor at rollout, with more coming online soon after.
And about food safety — the robot isn't wearing gloves in the video, but it will be at launch. Sharpa has designed a special pair of gloves so the robot and hands meet strict IP and food-safety standards.
If you can't make the launch, the operations will be long-streamed during business hours. Sweet.
More exclusive updates to come before the Shanghai launch. Be sure to follow.
Special thanks to CEO David Li and the @SharpaRobotics team for entrusting me with this exclusive first look.
Sharpa retweeted
👇🏽🔥🦾
We're excited about the launch to come! Thanks @GoingBallistic5 for highlighting the challenges of real store operations in F&B! #North is ready to take on the challenge🫡
Sharpa retweeted
How can robots learn dexterous manipulation from human demonstrations at scale?
Excited to share CHORD: Learning Dexterous Manipulation Using Contact Wrench Guidance From Human Demonstration.
CHORD learns from human demos by focusing not only on where contact happens, but how that contact moves the object through force and torque guidance.
This unified contact-wrench representation carries human manipulation skills across diverse behaviors, long-horizon tasks, whole-body embodiments, and real-world hardware.
We evaluated CHORD on large-scale, long-horizon, contact-rich tasks paired with human demonstrations, spanning rigid, articulated, and multi-object manipulation.
At scale:
* 82.12% average success across 1,831 tasks
* 90.77% whole-body manipulation success
* 4,739 sim-ready dexterous manipulation benchmark
* Transfer to real dexterous hands
Project page: nvidia-isaac.github.io/video…
Tech report: nvidia-isaac.github.io/video…
Code will be released soon as part of Video to Data repo github.com/nvidia-isaac/vide…, our end-to-end pipeline for converting human demonstration videos into simulation-ready assets and physics-grounded robot training data.
Huge thanks to amazing contributors: @zhu_xinghao , Zixi Liu, Shalin Jain, Chenran Li, Milad Noori, Huihua Zhao, John Welsh, @michaelv03, Wei Liu, @TingwuWang , Xingye (Dennis) Da, @zhengyiluo, Vishal Kulkarni, @sNaema, @yukez, @DrJimFan, @bowenwen_me, @danfei_xu, @SohaPouya, @Dr_YanChang.
#Robotics #PhysicalAI #DexterousManipulation #RobotLearning #NVIDIA