The Sim-to-Real Gap Shrinks: Tactile Policies, Fluid Physics, and Community Tools

Research

Tactile Manipulation Policy Learns Directly from Sensor Data

This paper shows you can learn complex manipulation policies directly from raw tactile sensor data without needing a separate state estimator. For builders, this simplifies the stack and gets you closer to a plug-and-play tactile sense for your agents.

Improved Sim-to-Real Transfer for Legged Locomotion via Domain Randomization

This research refines domain randomization techniques to make simulated legged robots move more reliably in the real world. It's another step toward closing the sim-to-real gap for dynamic tasks without massive real-world data collection.

Tools

NVIDIA Isaac Sim: New Physics Compiler for Fluids and Deformables

The latest update to Isaac Sim includes a new physics compiler that handles fluid and deformable object simulation more efficiently. This is a tactical win for builders tackling tasks involving liquids or soft materials, reducing the need for custom code.

Events

ROSCon 2025: Focus on Real-World Deployment and Tooling Maturity

The ROSCon schedule this year is heavy on talks about deploying robots in unstructured environments and improving developer tooling. It signals the community is shifting from pure research to shipping production-ready systems.