Boston Dynamics Killer? 6 NEW Autonomous AI Robot Features (GOOGLE OMNI)

Then it must identify which is the peach. And while it picks these up, it has to use the proper amount of force as well as its upgraded multimodal perception system, which has been integrated with a large-scale spatial perception foundation model. And they’re using Matrix’s proprietary wave physical intelligence model, and here you see it hands over that peach. Next, we move to feature number two, which is flexible manipulation and tool operation. You see a puddle of water has been poured. Now this robot has to determine where is the water, then it identifies that it’s going to use a soft cellulose sponge where it needs to form its fingers around the sponge, but with not too much pressure so as to completely condense the sponge and then not pick up the water.

This is using its fingertip sensor array that’s capable of detecting as little as 0.1 newtons of force. Then we have feature number three, which is embodied chain of thought after this human has sneezed without issuing a voice command. The robot determines that what is appropriate here is to locate the XYZ coordinates of this tissue, grab it with as little force as possible, pull it out and hand it to the human. Next, we have feature number four, sorting manipulation where the human simply points to the box that should be picked up without issuing a verbal command, and the robot is going to determine here to use just one hand to pick up this box, opening its fingers by about 30 degrees, using about 0.2 newtons of force so as not to crush the cardboard and then hand it to the human.

Then we have feature number five, which is bimanual coordination. This robot is going to pick up an egg and transfer it into its other hand, and we’re just going to fast-forward here, but the entire task takes about 30 seconds, and this is an example of its 3D woven biomimetic skin, where it has flexible fabric that’s covering the robot chassis that acts as a distributed sensing network to cushion the contact and detect impact forces in real time, such as this egg. And finally, we have feature number six, which is dual robot coordination.

So this robot is going to pass a box from one Matrix 3 to the other, which takes about a minute and 45 seconds, which we will fast-forward through, but in terms of price, this robot is going to cost $99,000 for outright buyers, and Matrix Robotics is also introducing a robot-as-a-service subscription framework for early access partners. And when it comes to their planned manufacturing scale, they’re looking at a capacity of about 10,000 units to start. So tell us down in the comments how long you think it’ll take for this robot to catch up to human level speed.

But Unitree is also entering the voice-driven arbitrary action space for humanoids as they just released this demo of their G1 humanoid listening to voice commands and carrying them out. This first one here it’s been told to jump in place and then do a plank, and you can see it performing this in real time as it computes its action trajectories. Then the woman tells the robot to turn around, put its hands on its hips, and wiggle its butt, presumably for the next TikTok trend. Here we can see the robot again computes its action trajectories and begins to dance, dance, dance.

And this entire video was shot by Unitree within one take. And the question here is beyond these fun little dance moves, what could this robot do if it was given dexterous hands? What is the future for these types of robots in, for instance, service industries? Or could there be military applications? Comment down below what you think could be the next frontier of voice-guided robotic action that would hopefully be monetizable and useful to someone. But an even bigger frontier than Earth is space as Swiss startup Orbit Robotics just officially announced Helios, which is a specialized humanoid platform designed strictly for microgravity and orbital servicing.

And it’s developed as a spin-off project out of East Zurich, where this robot breaks completely away from terrestrial humanoid designs by optimizing its entire mechanical architecture for zero-g environments. With four arms, because its legs serve no purpose for walking in space, the Helios features four symmetric arms that it can use interchangeably to navigate spacecraft interiors, hold structural handrails, and handle tasks simultaneously without needing leg anchors. Plus, it features tendon-driven actuation, which is to minimize moving mass and rotational inertia so that all the primary drive motors can sit deep within the core chassis near the shoulders, and then the power is transmitted down the limbs via high tensile spool and cable framework.

On top of this, it features rolling contact elbow joints, which are important because they replace traditional heavy gearboxes to ensure virtually friction-free motion while delivering a vital balance of structural strength and physical compliance, which allows the Helios to be able to absorb sudden orbital impacts or vibrations while it’s out in space without being damaged. But Gen AI also just had a breakthrough as Google just officially introduced Gemini Omni, which is its foundational world model that’s engineered to generate and edit high fidelity video by blending text, audio, images, and video inputs into a single system.

And Omni marks a massive shift in AI realism by embedding an intuitive grasp of physics like kinetic energy, gravity, and fluid dynamics directly into its generations, which allows it to translate complex concepts into accurate video sequences or even morph native video assets via simple conversational text commands. Plus, Google DeepMind also just expanded Project Genie, combining its generative world model with Google Maps colossal database of 280 billion Street View images spanning 110 countries, and they just announced it at Google I.O. 2026, allowing users to play 3D simulations that are anchored in real-world geography.

And it utilizes a new framework called Maps Imagery Grounding, where users can drop a pin anywhere in the United States and then command Genie 3 to render that exact location on the fly. Anyways, like and subscribe. Check out this video here for more of the latest in AI and robotics news, and thanks for watching!
[tr:trw].