D-Robotics detailed Uranus, a unified robot video generation model that goes from 2-second training clips to stable 60-second closed-loop generation. The engineering breakdown covers long-horizon generation, multi-camera support, and a single model covering 7 mainstream robot embodiments. The technical report, inference code, SDK and model weights are now public, and 100 early-access spots are open worldwide for developers and researchers.

A new grooved fiber actuator reported in Science Advances uses channel geometry to program bending and twisting, simplifying soft-robot control and expanding design options.
Mogan Technology is developing a whole-body electronic skin that moves tactile sensing beyond fingertips, with Zhiyuan and LimX Dynamics reportedly backing the effort. The approach aims to give robots richer physical perception for manipulation and interaction.
Two robots developed at the University of Tokyo can suspend a simulated patient in the air using their legs as manipulators, and can also reconfigure into a tow truck mode.
Tongji University researchers have developed a bio-inspired impulse arm for aerial robots that snaps open in 0.07 seconds, cutting impact forces by about 75% while preserving flight stability.
RoboParty, a startup less than a year old, has fully open-sourced its bipedal humanoid robot prototype, including mechanical drawings, BOM, motion algorithms, a troubleshooting database and supply-chain details. The move aims to lower development barriers and challenge the closed engineering model in humanoid robotics.
On October 27, an embodied AI evaluation benchmark led by the University of Hong Kong released its latest rankings, with Zhiyuan's model taking first place with an average score of 61.90 and a success rate of 56.42%, surpassing general large models. The results indicate a growing division of labor between general and embodied models, while the benchmark provides an objective capability standard for the industry.
Robbyant has closed the qualifier round of its Embodied AI Challenge, with 1,855 teams, 3,252 participants and 648 organizations taking part from 13 countries. The company described the result as a first milestone for the competition and said a 48-hour hackathon on real robots is coming next.

Xindu Qiyuan is promoting a third route for embodied intelligence that avoids relying on massive GPU clusters, according to Leaderobot. The company positions compute-efficient design and data-driven methods as an alternative to brute-force scaling.
AGIBOT’s GO-3 Nano Preview has taken first place on RoboDojo’s latest leaderboard with an average score of 61.90 and a 56.42% average success rate, surpassing GPT-6 Astra. It also ranks first in fine manipulation, long-horizon execution, and open semantic understanding.
Amap, Alibaba's map and navigation arm, has unveiled a whole-body foundation model that uses only human action data to help humanoid robots learn coordinated movement, posture and two-handed manipulation. In simulation and real-robot tests, the model reportedly leads comparable methods while collecting data about 3.4 times more efficiently.