Embodied Intelligence Advances with GPT-6 Astra's Robotics Testing and RPent Launch
- Published
- Source
- leaderobot.com
As GPT-6 Astra begins real-world robotic operations, the paradigm of intelligent agents is transitioning from the digital realm to physical applications. This shift raises a critical question: can general large models serve as the 'brain' for robots, enabling them to perform complex tasks in the real world? Physical agents must not only understand but also perceive, act, react quickly, and retain information.
The RPent infrastructure, initiated by Tsinghua University, Wuyuan Xinqiong, and Zhengxing Innovation, has been officially open-sourced. RPent connects the task understanding and planning capabilities of general large models with the fine operational skills of expert models like VLA, creating a complete feedback loop from perception to decision-making, execution, and error correction. This allows agents to adapt continuously to environmental changes and to accumulate verified experiences into reusable capabilities, achieving a 92.6% task success rate in the LIBERO-PRO benchmark and optimizing end-to-end task completion speed by over seven times.
The release includes a series of RPent real-machine demos showcasing various open-ended tasks. Notably, these tasks do not require separate strategies for each scenario. The significant change in robots is their shift from merely executing learned actions to understanding tasks, invoking capabilities, and adjusting actions based on environmental changes. The ability to reorganize existing skills to tackle new tasks is crucial as robots navigate an open world filled with new challenges.
Original report
What Does Embodied Intelligence Need as GPT-6 Astra Moves to Robotics?
leaderobot.com · LeaderobotThis briefing is an independently written summary based on publicly available reporting and is provided for industry information and news discovery. The original report and source publication are credited and linked where applicable. RobotToday does not claim ownership of third-party source material.
Rights concerns? Contact [email protected] with the relevant URL and details. We will review the matter and take appropriate action where warranted.
More related news
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- 10
- 11
- 12
Related Suppliers
Robocurve
San Francisco public benefit corporation founded 2026 that benchmarks frontier AI models on real robot arms and maintains the Inspect Robots framework.
Shenzhen Liangyuan Innovation Technology Co., Ltd. 深圳亮源新创科技有限公司
Shenzhen embodied-AI lab led by ex-OpenAI researcher Roger Jiang, building Physical AI foundation models Light-O1, Light REACT and LightNav-0
Striding.AI
Striding.AI (Striding AI) is a Beijing-based embodied-AI and robotics company founded in early 2026. It is led by founder and CEO Yao Song; co-founder Yu Chao works on reinforcement-learning post-training, and co-founder Yang Yuxin was named President for global expansion in September 2026. Funding is reported at nearly 100 million USD in total, with the latest angel round announced in June 2026 and investors including Charoen Pokphand Group, Huaqin Technology, and Jiuan Medical. The product line includes H1, a bipedal humanoid service robot; C1, a wheeled dual-arm restocking and transport robot; A1, a planned compact robot for high-precision handling of delicate items such as checkout tasks; and M1, a fleet platform for robot lifecycle management, orchestration, and optimization. On the software side it has developed SLM-0.5, a latent-space world action model with 2.3 billion parameters, the STEAM post-training system based on reinforcement learning, and the Rpent embodied agent framework. H1, C1, M1, and SLM-0.5 were launched in October 2026, and commercial service is planned for 2027. The company has said it will start with retail deployments such as shelf restocking and inventory counting.
Hitch Interactive
San Ramon, California embodied-AI firm that sells and rents AgiBot humanoids and runs HOPE, an official event of the 2026 World Humanoid Robot Games.
ACCESS
ACCESS Co., Ltd. is a Tokyo-listed (TSE:4813) Japanese software firm known for the NetFront browser, now testing VLA-model remote robot control over IOWN APN.
Humanoid
UK startup building commercially scalable, general-purpose humanoid robots via its KinetIQ VLM/VLA framework, offered as RaaS.
Ocean Infinity
Marine robotics company operating large uncrewed surface vessels and AUVs/ROVs for remote ocean survey and seabed operations.
Arkisys, Inc.
California company running NASA's Astrobee free-flying robots on the ISS under a Space Act Agreement and developing the robotic, modular Port orbital platform.
REVOBOTS
AI-driven 3D-printed task-adaptive humanoid robots for industrial workforce automation, offering TASKBOT AVA humanoid and TASKBOT SCOUT patrol platforms.
Flexiv Robotics 非夕
Santa Clara adaptive robotics company founded 2016; makes force-controlled Rizon arm series with 0.005N force sensing; raised $322M total.
Generalist AI
Frontier AI research company building embodied foundation models for general-purpose robots, with GEN-1 achieving 99% task success rates.
Dyna Robotics
US startup building embodied AI foundation models for dual-arm commercial robots, achieving 99%+ success in 24/7 deployments in restaurants and laundromats.
Robotics needs a service framework.
RSF defines a common language for robot service capability, lifecycle operations, certification pathways, and service-provider networks.
Join the RobotToday community on LinkedIn
Daily robotics news, in-depth analysis, conference highlights, and discussions with professionals worldwide.