Alloy: A Multi-Material Physics Engine for Robot Learning
AuthorsPranay Gupta, Deepanshu Rohilla
AffiliationsArgon Robotics
Alloy brings robots, cloth, cables, liquids, and other materials into one GPU-capable simulator for robot learning.
Key results
Reached after 15.5 minutes of training across 4096 worlds.
Reached after about 30 minutes of training.
Reached after about 70 minutes of training.
Approximate throughput increase from 1 to 64 parallel worlds.
Failures among 1024 worlds in the reliability test.
What the paper found
Alloy brings robots and nonrigid materials into one robot-learning simulation scene, automatically routing objects to suitable solvers: MuJoCo Warp for rigid bodies, Newton’s VBD for cloth and soft bodies, implicit MPM for granular materials, and SPH for liquids. Proxy bodies exchange poses and contact forces between solvers within each control step, while adaptive step sizes and per-world OK, DEGRADED, or INVALID health grades help contain unstable simulations. The engine also exposes tools through Anthropic’s Model Context Protocol, an interface tested by coding agents including Claude Opus 5.5 on an NVIDIA L4 GPU. In those cold-start trials, a Unitree Go2 walking policy reached 97% held-out success after 15.5 minutes, cable lifting reached 95.7% after about 30 minutes, and a custom cup-pouring policy achieved 100% after about 70 minutes. On the same GPU, Franka-scene throughput increased roughly 39 times when scaling from 1 to 64 parallel worlds; in a separate reliability test, 0 of 1024 worlds failed during a five-second block-tower simulation. Alloy’s central contribution is not a new solver, but an integrated engine that combines material-specific physics, coupling, diagnostics, and agent-facing tools in a shared, parallelizable workflow.
Original abstract
Robots handle much more than rigid objects: towels, cables, water, sand. The simulators used to train robots at scale are strongest on rigid bodies, and adding another material usually means a second engine and hand-written glue between the two. We present Alloy, an open-source physics engine for robot learning in which rigid robots, cloth, cables, liquids, granular material and soft bodies share one scene. Each object is described once; Alloy assigns it to a solver suited to its material, couples the solvers inside every control step, and runs the scene in many parallel worlds on one GPU, or on a CPU. Every world carries a health grade, to help tell a failure of the physics from a failure of the policy. Alloy is built on Newton, MuJoCo Warp and Warp, and it is written to be driven by coding agents: its documentation ships inside the package, every error names its fix, and the engine is exposed as tools over the Model Context Protocol. In a cold test, agents that had never seen Alloy installed a pre-release from the package index and trained policies for walking, cable handling and pouring on a single GPU. We also report how throughput grows with the number of worlds, depth-camera throughput, agreement with closed-form references, and a reliability test over a thousand worlds.
Read the original paperMore in Robotics
Browse all 68 papers →ELA: End-to-End Latent Action Policy Learning from Video Demonstrations
Cheng-Chun Hsu, Peter Stone
ELA turns plentiful robot videos into useful control experience by jointly learning latent actions, policies, and action decoders.
GRASP: Gradient-Guided Sampling for Contact-Rich Physics-Based Motion Retargeting
Yang Liu, Tong Wang, Yiming Bao, Heng Yang
GRASP uses past simulation rollouts to guide robot motion retargeting, producing contact-rich movements that work in simulation and on real hardware.
PERA: Plug-in Experts for Robotic Actions in Continual Whole-Body Control
Shangchao Hu, Yadan Sun, Xibo Yu, Wei Huang
PERA teaches humanoid robots new skills by plugging small experts into an existing controller without disrupting its original motions.