Search

UEMA: from a first touch to a growing self

The document outlines UEMA, a conscious-AI research project built on a simulated body (MIMo) with MuJoCo physics. It emphasizes rich, whole-body sensing—vision, touch, proprioception, vestibular, and hearing—integrated to enable progressive self-understanding and capabilities, using simulation to repeat experiments and connect multi-sensory evidence toward self-driven learning and behavior.

From the intelligence that analyzes to the intelligence that acts: the era of Physical AI - Tom's Hardware

Translated title. Original: Dall’intelligenza che analizza all’intelligenza che agisce: l’era della Physical AI - Tom's Hardware

Physical AI extends AI from digital outputs to real-world interaction, forming a closed Sense-Understand-Decide-Act loop with continuous feedback. It moves from perception to action, enabling machines to modify their environment in real time and transforming maintenance and human–machine collaboration, while facing latency and determinism challenges.

[2611.09847] Computer-Use Agents Are Embodied Generalists

The paper presents computer-use agents as generalists for embodied tasks, using software that maps physical environments to screen views and controls. A single agent design earned placements in four ECCV workshop competitions, leading the authors to argue that software skills can transfer to embodied AI.

The arrival of physical AI: when artificial intelligence leaves the screens | WIRED

Translated title. Original: La llegada de la physical AI: cuando la inteligencia artificial sale de las pantallas | WIRED

Physical AI moves AI from screens into the real world, enabling machines to sense, understand, and react to their environment with onboard hardware. It reshapes industries by pairing AI with physical bodies like robots and autonomous systems to operate with real-time perception and movement, reducing reliance on cloud processing.

Humanoid robots learn to imitate human actions - what is embodied AI

Translated title. Original: Гуманоидные роботы учатся повторять действия людей - что такое embodied AI

Embodied AI, which embeds AI into physical robots, is advancing as Chinese firms shift focus from hardware to intelligent control. Competition with US startups, large investments, and data-driven training methods enable robots to learn from real-world examples, but challenges remain in sim-to-real transfer and chip shortages.

Intelligent robots - Study: Germany leads in physical AI

Translated title. Original: Intelligente Roboter - Studie: Deutschland führend bei physischer KI

Germany leads in physical AI readiness and startup activity, with about a quarter of German startups involved and a favorable view of Germany as a location. Despite positive readiness perceptions, customer hesitancy slows growth, though most respondents expect physical AI to become standard within five years.

[2610.26765] HandsOn: A Large-Scale Simulation Benchmark for Generalizable Dexterous Manipulation

HandsOn is a composable simulation platform and benchmark for cross-embodiment dexterous manipulation, pairing 18,000 demonstrations with 66 tasks across Core-50 and Transfer-16 to study generalization, sim-to-real transfer, and limits of reliable physical interaction.

Embodied AI unveils wheeled AIR robot for cage-free factory work

Embodied AI has introduced AIR, a wheeled two-arm robot designed to work alongside factory staff. It says AIR can learn tasks from remote operators and reports 95% average success in paid pilots, but has not released key specifications, pricing, or a delivery date.

Intelligent Robots – Study: Germany Leading in Physical AI

Translated title. Original: Intelligente Roboter - Studie: Deutschland führend bei physischer KI

A German-led study finds Germany is strong in physical AI, with about a quarter of startups active in the field and high rankings in workforce, supply chains, capital, regulation, and market readiness; however, customer hesitation and perceived national readiness lag behind, with many predicting wider adoption within five years.

Google lance Gemini Robotics 2.5 : l'IA qui rend les robots plus habiles et plus précis

Google says Gemini Robotics 2.5 can learn hundreds of everyday tasks, including folding laundry, serving pastries, and sorting car parts on its own.