Agility outlines known security risks in robotics: back doors, unauthorized commands, and one robot's access being used to reach others. CTO Pras Velagapudi says the company designs against these threats, and explains why Digit verifies every exchange at both ends of a connection.

Runway has released Praxis-1, a 27-billion-parameter open-weight world action model for robotics, under Apache 2.0. The company says it is the top-ranked world action model on NVIDIA's RoboLab leaderboard and has adapted across humanoid, industrial arm and dual-arm warehouse robots.
Jiuzhiyang (300516.SZ) said on an investor interaction platform that its technical reserves in infrared imaging, laser ranging and other photoelectric perception sensors can be applied to robotic platforms.
Hang Zhao, assistant professor at Tsinghua and co-founder and chief scientist of Galaxea, delivered one of this year's Early Career Keynotes at CoRL 2026. His argument: the physical world does not pause while a robot computes. Headline numbers look strong, but the bottleneck lies elsewhere, with inference latency rather than the controller setting the pace. Zhao therefore splits robot intelligence by timescale instead of by perception, planning and control, from System 2 reasoning over seconds down to fast System 1 policies at tens of cycles per second.

At CoRL 2026, RoboSense and BitRobot announced a strategic partnership on 3D perception and real-world data collection. Their first public result is BitRobot's RoboCap LiDAR-100 dataset, containing 100 hours of multimodal kitchen, auto-repair, and car-detailing operations.
On November 11 during CoRL 2026, RoboSense announced a strategic partnership with BitRobot focused on 3D perception and real-world data collection. As an early result, BitRobot released the RoboCap LiDAR-100 dataset, built on RoboSense's 3D perception system and covering 100 hours of real kitchen, auto-repair and car-detailing work.
Sharpa co-founder Dr. David Li appeared on CNN's Global Perspectives in Doha to discuss the future of robotics, covering general-purpose intelligence, dexterous manipulation and how robots can boost human productivity. Sharpa's stated view: one shared model, dataset and physical architecture will let robots learn across many tasks, and dexterous hands are the key.
Sharpa co-founder Dr. David Li appeared on CNN's Global Perspectives in Doha to discuss the future of robotics, from general-purpose intelligence and dexterous manipulation to how robots can boost human productivity. He laid out four beliefs about the next era of robots, including that one shared model, dataset and physical architecture will let robots learn across many different tasks, and that the dexterous hand is the key.
A sub-1-gram flapping-wing robot has achieved untethered altitude-hold flight for four minutes, according to Leaderobot. The milestone addresses a key challenge in micro aerial robotics.
The 2026 AI100 report, the third in a five-year series, assesses AI’s technical and societal trajectory. It identifies physical AI and embodied intelligence as a key grand challenge while noting China’s dominance in industrial and humanoid robots and the persistent gap between demonstration and deployment.
Memories.ai is building a standardized egocentric video pipeline that pairs modular capture hardware with AI annotation to create robot-training data. The San Francisco company aims to become a long-term data-infrastructure partner for robotics and physical AI, using NVIDIA Cosmos to interpret and post-train on customer video.
Times Electric said Shanghai Aisendi has developed a heavy-duty subsea chain trenching and cable-laying robot for hard-seabed operations. It is intended for offshore wind and marine communications construction.
Topstar said on Nov. 11 that it has developed the X5 intelligent robot control platform based on a domestic operating system and built full-stack capabilities across core components, robot bodies and system applications.
AGIBOT has released LIFE, a lifelong embodied intelligence framework, and opened its RoboBeats model and evaluation resources to support standardized research.
Mech-Mind Robotics says its embodied AI "Eye-Brain-Hand" system tackles deep-bin random picking, a common bottleneck in real-world robot applications where complex part geometries and harsh environments complicate perception and grasping. The company reports a part pose recognition success rate of at least 99.99%, picking accuracy of ±0.3 mm, and bin emptying rates above 99.99%.
