Chinese researchers just taught a Unitree G1 humanoid robot to play tennis using only five hours of imperfect motion-capture data. Through a framework called LATENT, the robot can track footwork, return balls traveling over 15 m/s with 90% accuracy, and sustain multi-shot rallies against human players.
Nosh Robotics just unveiled Nosh One, a $1,499 AI kitchen robot that automatically adds ingredients, stirs, and monitors cooking via an onboard camera. The companion app lets you prompt custom recipes alongside 500 built-in meals, with Kickstarter preorders set to ship in summer 2026.
Figure just demonstrated Helix 02 cleaning a living room using a single end-to-end neural system running directly from pixels. Without changing algorithms or writing custom controllers, the robot learned complex tasks like spraying and wiping tables, tucking bins under its arm, and operating a remote simply by training on more data.
Researchers just open-sourced OmniXtreme, a control framework that teaches humanoid robots high-agility moves like backflips, breakdancing, and handstands. A two-stage approach combines scalable simulation pre-training with residual learning to protect real-world hardware. Why it matters: It breaks the generality barrier, proving humanoids can execute complex acrobatics with a single unified policy without wrecking their motors.
HONOR just revealed its Robot Phone concept and first humanoid robot at MWC 2026. The futuristic handset packs a micro 4DoF gimbal system and a 200MP camera, moving physically to nod, track subjects, and express emotion. Why it matters: Hardware makers are pushing to move AI out of glass screens and into embodied physical interaction.
Alphabet's robotics software bet Intrinsic just merged into Google as a distinct group, teaming up with Gemini, Google Cloud, and DeepMind. The move connects its modular Flowstate robotics platform directly to Google's frontier AI models to automate factory floors and warehouses. Physical AI is finally getting the Android treatment.
Unitree just revealed the As2, a quadruped companion robot packing 90N·m of torque and industrial-grade specs into a compact frame. It features over 4 hours of no-load runtime, IP54 rainproofing, and can haul a 15kg payload across more than 13km. With open secondary development, it basically bridges domestic companionship with practical utility.
Figure just put a fleet of seven Figure 03 humanoids into 24/7 autonomous operation at its California headquarters, powered by its Helix 02 neural network. The bots monitor their own battery telemetry, stepping onto 2 kW inductive charging docks and swapping tasks without human intervention. It marks a serious technical milestone for humanoid uptime, even if Figure hasn't disclosed what actual work they're doing yet.
A software engineer building a custom gamepad app just accidentally gained access to nearly 7,000 DJI Romo robot vacuums across 24 countries. Using an AI coding assistant to reverse-engineer cloud communications, he exposed live camera feeds and home floor plans. DJI rolled out an automatic patch on February 10.
OpenAI is quietly building a 200-person hardware team led by Apple veterans to launch consumer AI devices. Its first product is reportedly a $200 to $300 smart speaker packed with cameras for facial recognition and proactive life nudges, slated for February 2027 at the earliest. Smart glasses and a lamp are also in the pipeline.
Toyota just contracted seven Digit humanoid robots from Agility Robotics to work at its Canadian RAV4 plant under a robots-as-a-service deal. Following a year-long pilot, the bipedal bots will start unloading auto parts from tuggers this April. While backflip demos grab headlines, integrating real humanoids into daily factory shifts is a far bigger milestone.
Cyborg Dynamics and Griffith University just tested an AI-powered robot swarm that put out fires with a 99.67% success rate. Using multi-agent reinforcement learning across physical and virtual ground vehicles, the system self-organized to navigate obstacles and divide tasks. Why it matters: Autonomous swarming could soon replace manual remote controls and keep first responders out of extreme danger.
Unitree just revealed behind-the-scenes details of its Lunar New Year showcase, where dozens of G1 humanoids performed the world's first fully autonomous cluster kung fu routine with zero teleoperation. The demo also featured an H2 robot dressed as the Monkey King riding quadruped "clouds," proving multi-humanoid physical coordination is advancing rapidly.
Nvidia and university researchers just released DreamDojo, a robot world model trained on 44,000 hours of egocentric human video. Instead of collecting endless, costly physical demonstrations, robots learn basic physical intuition by watching humans before fine-tuning for specific hardware. It also runs in real time at 10 FPS, making pre-deployment simulations far easier.
Engineer Cyrus Clarke just brought the OpenClaw AI into the physical world using MIT Media Lab’s neoFORM shape display. Inspired by the project's name, the setup gives the agent a tactile lobster-claw form with custom physical body language and motion feedback. It offers a fun glimpse into how desktop AI might break out of chat boxes and into physical space.
NVIDIA just introduced Cosmos Policy, a robot control model built by fine-tuning its Cosmos Predict world foundation model. Instead of patching separate action modules onto vision-language models, it treats robot actions and physical states as latent video frames inside a unified diffusion process. It scored a record 98.5% success rate on the LIBERO benchmark, with weights and code now open on Hugging Face.
