Warehouse Ops, Reward Models, and the Omniverse Toolkit

News

Efficient Reward Modeling from Human Preferences

A method for training reward models using weak, middle, and strong human preferences shows promise for improving policy learning in complex robotic tasks. This is a practical step toward making human feedback more efficient for robotics applications.

Agility Robotics' Digit Enters Real Warehouse Operations

Agility Robotics announced that its Digit humanoid is now operating in a real GXO warehouse, performing tasks like moving totes. This is a tangible deployment in a major 3PL, moving beyond controlled demos into actual logistics operations.

Tools

NVIDIA Drops New Omniverse Code for Robotic Simulation

NVIDIA released new Omniverse code for developers, enhancing tools for robot simulation and digital twin creation. These updates are critical for builders simulating complex robotic interactions and testing in virtual environments before real-world deployment.