Alloy physics engine for robot learning Argon Robotics released Alloy, an open-source multi-material, parallel physics engine for robot learning built on Newton and MuJoCo Warp and licensed under Apache-2.0, installable via `pip install alloy-engine` on Python 3.11 or 3.12. Alloy steps robots, cloth, cables, liquids, sand and soft bodies together in one scene, runs the same script on an NVIDIA GPU or CPU, and ships with 5,423 library objects, 9 ready-made tasks and quadruped walking, plus a sim.health() check that distinguishes physics failures from policy failures. The engine is aimed at AI agents building training tasks from in-package docs, which Argon Robotics says agents used to build new tasks in minutes. A multi-material, parallel physics engine for robot learning. Built on Newton and MuJoCo Warp. Open source under Apache-2.0 https://github.com/argonrobotics-ai/alloy/blob/main/LICENSE . You pip install alloy-engine Your agent Build and train