NVIDIA Powered Humanoid Robot Tries Autonomous PC Build Plus This (AI NEWS)

➡ Robots are becoming more advanced, with abilities to assemble computers, peel apples, play table tennis, and even tennis. They use a combination of tactile sensors and cameras to understand pressure and movement. Some robots can also create real-time 3D maps of their surroundings to avoid obstacles. In the future, these robots could be used in various environments, from factories to homes, potentially replacing some human tasks.

Are skilled trades about to be automated too? Today on AI News we’re taking a closer look at several new humanoid tech demos for you to decide whether humans are about to have more help or more competition. Starting with this demonstration of the Sharper North robot assembling a computer using two sharper wave 22 degree of freedom hands and something to note here is that each finger features 64 sensing and it has a visual tactile setup here watch how it seats this GPU inside it has to have just the perfect amount of force here it does this using 1000 tactile sensors per fingertip which sense as little as 0.005 newtons each and it’s actually also using a mixture of experts approach with specialist neural networks that activate depending on the task phase like here grabbing this screw with the screwdriver and it actually feels by seeing by using the combination of both its tactile sensing arrays in the fingertips as well as its cameras which can actually see the fingertip deformations to give the robot a better understanding of its pressure to deliver the screw into the hole like this next watch how it approaches this plug that it’s about to pick up here and it’s actually operating at four hertz which means it refreshes four times per second and the question is what else would you want this robot to be able to do in order to justify a purchase here i’m going to show you another demo of the robot peeling an apple and this is thanks to the robot’s force guided tool manipulation also watch the way that it rotates the apple in its hand and it’s important to note that the apple peeling task only succeeded 30 percent of the time but the peel completion ratio hit 73 meaning that even failed attempts got most of the way through the sequence but what else would you want this robot to do and how much would you pay so far it can play table tennis as well as fold pieces of paper but a team of researchers are taking table tennis even further to actual tennis you’re seeing here unitries g1 being used in latent this stands for learning athletic humanoid tennis skills from imperfect human motion data so instead of needing perfect complete match data this system learns from short imperfect motion fragments of individual skills like these forehand strokes backhand strokes lateral shuffles and crossover steps but let’s watch this sustained long horizon rally with a human this is actually able to return balls traveling at 15 meters per second or 33 miles per hour and the team actually use dynamics randomization and observation noise in simulation to bridge the gap to the real world so the policy works on actual hardware without needing precise system identification of things like ball bounce or air drag and you can see just how good this robot is from the angle behind it watch how it negotiates its footsteps to be able to adjust in real time and to take it even further they even had these robots self-play in simulation but it would be interesting to see what other sports these robots can play and is this the future of sports betting next limbx dynamics just revealed its ollie robot with an upgraded awareness of its surroundings thanks in part to real sense ai depth cameras powered by nvidia’s cu v slam and this stands for cuda accelerated visual simultaneous localization and mapping here you can see this robot is building a real-time 3d map of its surroundings and it’s specifically optimized for nvidia’s jets and edge computing platforms to provide real-time low latency pose estimation and mapping as you can see here and as it’s walking through it will also see different obstacles to avoid even if they’re semi-transparent and here it is avoiding some more obstacles here so the question is what else can this robot do and how can this awareness be applied to the real world to either make money or serve as entertainment and the agi bot world challenge at icra 2026 is offering a 530 000 prize for a reasoning to action challenge where online simulation to real robot testing on the agi bot g2 will focus on robust physical interactions in complex environments and the developer resources include the agi bot world open data set genie sim 3.0 as well as an action chain of thought vision language action model and at gtc 2026 skilled ai just made one of the biggest partnership announcements in industrial robotics teaming with abb robotics universal robots and nvidia to deploy its foundation model called skilled brain across real factory floors at scale so the core idea here is that they call this omni-bodied intelligence which is one ai brain that can control any robot on any task without needing engineers to hand program each new workflow from scratch and the traditional approach to industrial automation was expensive and rigid because every task required custom code written by human experts which didn’t scale well especially for small and medium-sized businesses where products can change regularly so skilled approach instead flips this by using an end-to-end neural network that learns directly from data so for any new task the base model just gets fine tuned with a small amount of task specific data in a process they call post-training and this means manufacturers can extend automation into more dynamic and complex applications without rebuilding from the ground up each time and the partnership here is already showing real results working with nvidia and foxcon skilled is shipping its brain to control dual robotic arms on nvidia’s blackwell gpu production lines and the task involves picking and placing a bus bar positioning a limit block drilling 16 screws in sequence and removing the block which is a long horizon precision heavy job currently done by humans but this system handles it end-to-end with no step-by-step programming and can even recover on the fly when something shifts unexpectedly during the assembly process and there’s a clever data strategy at play here where skill describes a self-sustaining flywheel robots deployed in structured factory settings generate data that trains better models which then enable deployment in less structured environments like hospitals or hotels and eventually homes anyways like and subscribe and check out this video here for more of the latest in ai and robotics news
[tr:trw].