An interesting grasping project called HUG from NYU
the hand is the bridge and entrance of Physics AI--so,their idea is the best teacher for robot grasping is human demonstrations.
Using smart glasses, they collected ~28 hours of first-person egocentric videos to build the 1M-HUGs dataset from real human priors--6,700+ everyday objects across 41 buildings, with zero robot data.
and they use flow matching (learns smooth paths from noise to realistic grasps, faster than diffusion), it generates human-like poses from one RGB-D image and zero-shot transfers to robot hands.
Real-world home tests reached 62% open-loop success…
Temporarily set aside the fully humanoid
Genesis AI just launched Eno,their first general-purpose robot:
elegant, approachable, and pragmatism
Headless,wheeled base with an adjustable tower that rises to working height and folds compact like a carry-on. Worth highlighting is the custom dexterous hand: 20 active back-drivable DoF, sized and tuned for human-like manipulation and soft contact.
Powered by the GENE-26.5 foundation model, it handles rich, long-horizon tasks like cooking, lab work, cable assembly, and more…
---
Spanning data, models, and general-purpose robot, Genesis AI is accelerating the flywheel of physical AI through full-stack integration.
Seeing the ultra-realistic humanoid robot Moya casually walking into a human home,
it still feels like a scene from a sci-fi movie…
At 1.65m and 30kg, with 40 facial muscle units, a full tendon-cable drive system, over 100 degrees of freedom, warm silicone skin (32–36°C), full-body tactile sensing, voice interaction, and that DroidSoul engine giving it emotional perception like an amygdala--it’s designed as a true companion for home, elderly Care, office, or hotel…
What does it really mean when humanoid robots start feeling this… human?
World-model-driven Curr-0 features a clear tri-system architecture:
>System 2: Reasoning and vision-language processing layer
>System 1: Whole-body motion and stability
>System 0: Specialized in dexterous hand-object interaction
Note: It uses the 21-DoF @wuji_global Wuji Hand (you guys are really everywhere 😂).
The Human-in-the-World Model (powered by HumanEx data) enables immediate deployment and rapid iteration for Curr-0--a smart way to close the sim-to-real gap.
I think this tech report is worth a close read, the team has incorporated many considerations based on real-world tasks,specifically, the transfer of human experience to humanoid robots,which is incredibly valuable for helping robot models understand the physical world.
https://t.co/iLiICwxfqG
It’s truly impressive--humanoid performing dexterous tasks while on the move.
Shanghai-based Current Robotics released Curr-0: a humanoid foundation system for loco-dexterous manipulation. It runs on a single unified policy trained with over 21k hours of real human whole-body data on ~70 DoF hardware. The whole-body coordination in dynamic tasks looks remarkably natural.
Yeah,that could be quite relevant for caregiving, home-use, office assistant,complex task scenarios…
Interestingly, they also developed HumanEx, a wearable system for capturing complex whole-body and dexterous human demonstrations.
This goes beyond typical egocentric or UMI datasets by including whole-body coordination data -- an often overlooked aspect, since complex real-world tasks require not just hand dexterity but integrated arm-torso-leg movements.(We humans do the same thing,lol)
Similar to how Unitree G1 learns to use its arms and body to open doors when hands are occupied, this whole-body approach feels key for practical applications.
Humanoid robots just picked up another sport 🏓
AGIBOT A3 is now autonomously playing table tennis against humans--and it looks impressively smooth.
Developed with Peking University’s SpikePingPong algorithm and Huang Tiejun’s 20kHz high-frequency pulse camera, its vision response is 10x faster, enabling millimeter-level prediction and millisecond decisions for continuous rallies, trajectory tracking, whole-body planning, and seamless attack-defense switches.
This level of high-speed dynamic control is exactly what real-world applications need: like the safer human-robot collaboration in factories,responsive service/elder care robots…
Huge step forward. BAAI Conference 2026 delivering again.
Tencent released an open-source project: Hy-Embodied-0.5-VLA--an engineering-driven embodied AI system focused on real-world robot learning
With 10k+ hours of high-precision egocentric UMI data, strong innovations in cross-embodiment transfer and offline RL (FlowPRO), it goes smoothly from data to reliable hardware deployment.
Solid performance on RoboTwin benchmarks for practical tasks like hanging mugs, stacking, and cabinet operations.
--
(For LLM companies, developing robot models has become a necessity rather than an option)
Seres has deployed bipedal and wheeled humanoid robots at its Chongqing "Super Factory" for tasks such as guided tours, inspections, and quality control;
the facility is equipped with over 3,000 robots, achieving near-total automation.
Humanoid robots are effectively filling the gap in the workforce for unstructured tasks.
...and form a massive automated collaborative network with robots.
NVIDIA MotionBricks brings humanoid robots closer to natural, fluid movement
The system combines a modular latent generative model with Smart Primitives, packing more than 350,000 motion skills into a single backbone. It runs in real time at up to 15,000 FPS with just 2ms latency.
Navigation, object interaction, sitting, jumping, obstacle crossing, and even different movement styles can be composed zero-shot, like reusable motion modules.
It is already being used for whole-body control on the Unitree G1 and serves as the motion layer for NVIDIA GR00T Whole-Body Control.
What makes this interesting is the move from training individual behaviors toward building a shared library of motion capabilities.
If that approach scales, robot movement could become more modular, transferable, and easier to deploy across different tasks and environments.
Another adorable home semi-humanoid robot
From SF's Weave Robotics
2026 vs. 2025
the 2026 model more like a general-purpose product,feels beautifully refined--clean lines, soft rounded shapes, and that fresh green color palette that gives off such a warm, approachable vibe.
folding,tidying up, organizing, and sorting…
And i think it can expanding into even more everyday scenarios: hotel,nursing homes, supermarkets, and beyond…
Hard to say no to a cute little one
It’s only 12kg--like a toddler under 2,yet it has 21 joints and can run, jump, and gently hug you…
Beijing Luvbotics is redefining what a living humanoid robot, like a family member,while it certainly doesn't cook
, laundry,cleaning… but it's a real emotional companion.
>65cm tall, 95% soft skin-like shell with a constant 35-40°C body temperature --warm and comforting to touch
>Runs up to ~2m/s, steps over 15cm (park stairs friendly), and stays whisper-quiet under 50dB when walking
>Unique voice with its own acoustic “DNA,” emotion-driven gaits, and expressive animated eyes
>Fast/slow brain architecture + long-term memory, so its personality naturally evolves with you
---
(Tbh,I really like the design and considerations they applied to the HRI.)
A good use case:
Two desktop humanoid robots autonomously collaborating
on repetitive and tedious tasks
Maybe these can be handled by them in the future.
---
Finland’s MAKIINA is slashing physical AI costs with a Lego-style modular platform.
Impressive, with a BOM of less than $1500.
This likely indicates they are building a full in-house stack--motors, actuators, sensors, power electronics, and their own AI foundation models.
K-Pop dance by a bipedal humanoid robot from South Korea's ROBOTIS.
Learning from videos of humans.
It looks like it’s ready to go head-to-head with the Unitree G1 and AGIBOT X2.