OSRA and ROS community advance Physical AI and real-time generalist robotics at ROSCon 2025 in Singapore

Singapore — October 27, 2025 — Open Robotics, through its initiative, the Open Source Robotics Alliance (OSRA), today announced new community-driven initiatives at ROSCon 2025 to accelerate real-time robotics and usher in the era of Physical AI.

Key efforts include the Physical AI Special Interest Group (SIG), progress on the ROS Lyrical Luth release, and making ROS both accelerator-ready and accelerator-agnostic, with NVIDIA supporting these efforts and contributing core technologies such as GPU-accelerated abstractions and real-time execution frameworks into ROS.

“As robots powered by Physical AI move from labs to production, OSRA and the ROS community are hardening the stack for reliable performance, from Lyrical’s core improvements to accelerator-ready pipelines so that developers can deliver safe, responsive autonomy on any compute,” said Geoff Biggs, CTO of Open Robotics.

“To scale the next generation of robots, developers need a robust, production-ready software foundation that ensures real-time performance and a trusted path from virtual training to real-world deployment.” said Deepu Talla, vice president of robotics and edge AI at NVIDIA. “NVIDIA is contributing GPU-accelerated computing and simulation technologies to the Open Source Robotics Alliance to help build that accelerator-ready foundation directly within ROS”

The Physical AI SIG, chartered by OSRA’s Technical Governance Committee, is developing open standards for interfaces, data pipelines, reference platforms, and embodied AI to accelerate innovation by leveraging existing standards and tools already available in the  ROS ecosystem. Meetings are open for community participation via the Open Robotics calendar.

In addition to working to integrate AI-based workflows,  ROS Lyrical Luth development is also advancing, with themes of expanding accessibility and ease of use, as well as strengthening professional-grade development capabilities. The team is targeting a May 2026 release. In addition, the community is extending multi-vendor hardware acceleration through REP-driven conventions and zero-copy pipelines, enabling ROS graphs to scale across CPUs, GPUs, and FPGAs with improved throughput and latency.

These initiatives are the result of collective contributions under OSRA’s open, vendor-neutral governance. Notable contributors include:

• • Intrinsic: With core developers contributing across the project, Intrinsic is investing in improvements to grow the communities that can be involved including improving support for Rust, Bazel, and projects using Protobuf. And in recognition of the importance of AI, their team is supporting the SIG for Physical AI.

• • NVIDIA: Contributing to ROS 2 by advancing real-time execution, enabling accelerator-ready performance, and streamlining sim-to-real deployment. Additional contributions include engaging in the Physical AI SIG, authoring design guides and developing GPU-accelerated libraries and open-source developer tools such as Greenwave Monitor..

• • Robotec.ai: Leading the development of new standardized interfaces for simulators, enabling tighter integration between ROS and any simulator supporting those interfaces, and finer control over simulation for reinforcement learning.

• • Ekumen, BMW, and Huawei, who are leading working groups under the SIG on standardising interfaces, training and execution pipelines, and data collection pipelines, respectively.

 

 

OSRA invites developers, startups, industry leaders, and researchers to join in shaping Physical AI and real-time ROS through contributions, Supporting Individual membership, or consortium membership with technical governance representation. The Physical AI SIG is now onboarding contributors.

About ROSCon 2025
ROSCon is organized by Open Robotics and is the leading open source developers conference for the ROS and robotics community.  Held annually since 2012 in various locations worldwide, ROSCon will be in Singapore from October 27-29 this year.