A new robot can navigate through 5cm diameter pipes with sharp turns while towing up to eight times its own weight, enabling versatile inspection in complex environments.
A University of Tokyo team used reinforcement learning to enable an inverted bipedal robot to perform 12 consecutive giant swings on a real high bar, demonstrating sim-to-real motion control.
A new cover review drawing on more than 100 studies compares soft robotic actuation technologies for implantable, surgical, and wearable applications, offering guidance on selecting the right artificial muscle.
MIT researchers have developed a bee-inspired micro robot that can fly, but it still relies on external power; an onboard battery is the missing piece for autonomous pollination.
Booster Robotics has spotlighted its joint system, saying it is built for power and tested for endurance. The company said the hardware is engineered to deliver powerful performance and lasting reliability, from lab validation through to real-world applications. No specific figures or deployment timelines were disclosed in the post.

Boston Dynamics congratulated the University of Maryland's Team RoboScout on winning the DARPA Triage Challenge. The company said it was incredible and inspiring to see tools like its Spot quadruped used to help emergency workers locate victims, assess injuries and prioritize lifesaving care.
Robot Talk Episode 171 features Sebastian Castro discussing open-source software for robot task planning and behavior development. The conversation covers manipulation, accessible tools, and Castro's work across NASA, PickNik, Boston Dynamics, MIT CSAIL, and robotics education.
A new adaptive AI agent is reported to connect experimental planning with robotic control in a wastewater treatment test. The work illustrates efforts to move research agents beyond software-only analysis toward physical experiment execution.
Leaderobot reports that Ego2X extends egocentric video beyond pretraining by converting human manipulation demonstrations into usable robot training data.
Mifeng Technology released a real-robot data quality study for embodied AI, finding that demonstration inefficiencies persist in post-trained models and that complete observations can sharply improve decision accuracy.
Mifeng has released a 200,000-hour real-world robot data quality study focused on what makes post-training experience effective for embodied agents. The work examines data value beyond volume as embodied AI data sources multiply.
D-Robotics has opened its Frontier technical webinar series to robotics researchers, students and engineers working on reinforcement learning, world models and learning-based control for humanoid robots. The series covers cutting-edge robotics research, advanced algorithms and emerging technologies shaping the field. The company invites the community to follow the series, share feedback and suggest topics or research directions they want to explore together.
On November 19, 2026, Qianxun SI's robot dog completed a last-mile delivery training exercise on Fuxing Island, Shanghai, showcasing integrated indoor-outdoor navigation and cross-floor operation capabilities.