Factory / IndustReal

NVIDIA's contact-rich assembly simulation and transfer pipeline: simulate tasks like nut-threading fast, then transfer the policy to a real robot.

Factory is a contact-rich simulation method and learning toolkit NVIDIA published at RSS 2022: it uses a signed distance field (SDF, which records how far every point is from an object's surface) for collision detection, combined with contact-point reduction and a Gauss-Seidel solver, to simulate 1,000 nut-and-bolt assemblies in real time on a single GPU, and ships 60 part models, 3 assembly environments, and 7 controllers, integrated into both PhysX and Isaac Gym. IndustReal (RSS 2023) builds on this, training reinforcement-learning policies for peg and gear assembly with techniques such as an SDF-based reward, a sampling-based curriculum, and a policy-level action integrator, reaching 83–99% success over 600 real-robot trials, and open-sourcing a 3D-printable parts kit along with deployment code. Together they show that high-precision assembly can also follow the “train in simulation, transfer to the real robot” path.

ExampleThe Isaac-Factory-PegInsert-Direct-v0 (peg insertion), GearMesh (gear meshing), and NutThread (nut threading) environments in Isaac Lab all come from Factory and can be trained directly with PPO.

Also called

Fast Contact for Robotic Assembly, IndustRealKit, IndustRealLib

Related

Contact-rich Manipulation · Peg-in-Hole Insertion · Sim-to-Real Transfer · NVIDIA Isaac Lab · PhysX · Signed Distance Field / Function

Sources

Factory: Fast Contact for Robotic Assembly (arXiv 2205.03532)
IndustReal: Transferring Contact-Rich Assembly Tasks from Simulation to Reality (arXiv 2305.17110)
Isaac Lab Available Environments

As of

2026-09