Robotics & Physical AI

San Francisco startup Tau Robotics is testing a humanoid cleaning service priced at about $30 per hour. The robot can mop, wipe surfaces and take out trash, but today it relies on remote human control.

ACE-Data-0 targets a central bottleneck in embodied intelligence: the lack of synchronized, human-centric data that connects perception, motion, manipulation, objects, sound and touch over time.

Google DeepMind has introduced Gemini Robotics ER 2, a high-level embodied reasoning model for robots. The release focuses on real-time video understanding, tool orchestration, and collaboration across multiple robot types.

TurboVLA challenges the common LLM-centric design of vision-language-action models by mapping vision and language more directly to robot actions. The paper reports 32Hz inference on an RTX 4090 with only 0.9GB VRAM.

Robotics companies are running into a shortage of high-fidelity real-world training data. Encord is testing whether brain-wave signals, combined with egocentric video and dense annotation, can help train more capable robot models.

NeoteAI and Fudan University have released three N0 technical reports focused on tactile data, vision-touch-language-action models, and tactile world models. The message is clear: robots need more than vision to operate reliably in the physical world.

Open-AoE is more than a dataset release. It packages collection, structuring, and downstream adaptation into an open pipeline for embodied learning.

HarmoHOI tackles a core bottleneck in hand-object interaction synthesis: keeping multiple views visually consistent while preserving plausible 3D motion. It does so by jointly modeling RGB videos and aligned 3D point tracks in one diffusion framework.

RynnBrain 1.1 is built to connect perception, spatial reasoning, localization, and planning in one embodied framework. The key shift is not just better understanding, but outputs that are more directly useful for robot manipulation.

JoyNexus proposes a service-oriented framework for post-training Vision-Language-Action models, aiming to make supervised fine-tuning, reinforcement learning, and evaluation more efficient under shared infrastructure.

A KAIST AI paper introduces robot-centric pointmaps to reduce the mismatch between camera observations and robot-frame actions in VLA models. The method encodes 3D scene coordinates in the robot frame while keeping the image-like grid structure used by existing 2D vision backbones.

Xiaomi-Robotics-1 explores whether robotics can benefit from the same scaling dynamics that transformed language and vision models. By pre-training on more than 100,000 hours of real manipulation trajectories, the model shows consistent gains as data and model size increase.

Jensen Huang’s two-day visit to Tokyo tied together Japan’s compute ambitions, robotics sector, auto industry, and chip supply chain. Nvidia is positioning itself not just as a GPU vendor, but as the infrastructure layer for Japan’s physical AI push.

Riemann Dynamics has introduced Riemann-1.0, a robot manipulation model that reached a 62.6% success rate on RoboCasa-365. The more important story is its training recipe: large-scale human egocentric video pretraining followed by robot action alignment.

RoboTTT brings test-time training into robot foundation models, enabling policies to condition on up to 8K visuomotor timesteps without increasing inference latency.

Agility Robotics is opening a 60,000-square-foot training facility in Fremont, close to where Tesla is expected to build Optimus. The move highlights a split between ambitious humanoid visions and robots already working in warehouses and factories.

Hyundai’s plan to deploy Boston Dynamics’ Atlas robots has triggered a strike at its Ulsan plant. The dispute shows how humanoid automation is becoming a concrete labor issue rather than a distant technology story.

A new arXiv paper introduces SPINE, an agentic framework designed to debug and deploy bimanual robots with less reliance on robotics experts. The work targets the often overlooked integration layer between AI “brains” and physical machines.

A new arXiv paper introduces SVR-MARL, a framework that evaluates the task value of sensed information under realistic underwater communication and covert-operation constraints.

A new arXiv paper proposes Recursive ArUco markers, a fiducial marker design intended to help UAVs recognize landing pads from both distant and close viewpoints. The approach embeds complete markers inside the black and white bit regions of a parent marker, enabling recursive detection without relying on the center being visible.