LLM-Coded Grasps: Shipping Robot Capabilities Without the Bottleneck

Research

No More Endless Prompting: LLMs Directly Write Robot Code from Language

This paper is a serious leverage point for builders. It cuts through the entire mess of formalizing robotic tasks by using a large language model to directly translate natural language instructions into robot-executable code, a massive reduction in time and expertise needed to ship a new capability.

Universal Dexterous Policy: One Policy to Grasp Them All

The sim-to-real tax is getting slashed. This research shows how to train a single, universal dexterous policy in simulation that works on a variety of different robot hands in the real world, a massive win for portability and reducing per-platform engineering debt.

Community

ROSCon 2025 Video Library: Your Cheat Sheet for OSS Robotics

ROSCon 2025 sessions are now posted, which is your cheat sheet for where the open-source robotics ecosystem is heading. For builders, this is salvage material for finding the next critical middleware stack, the emerging standard for interoperability, and the next potential hire who just gave a talk.

Tools

NVIDIA Isaac Sim Gets Another Performance Upgrade

NVIDIA keeps tightening the sim-to-real loop with another set of developer-focused updates to Isaac Sim. This release focuses on improving performance and making it easier to integrate custom assets, so you can generate higher-quality synthetic data much faster.