NVIDIA & UW Introduce Factory: A Set of Physics Simulation Methods and Learning Tools for Contact-Rich Robotic Assembly

When we think of modern industrial assembly lines, we can imagine a tireless ensemble of task-specific robots efficiently cutting, stamping, connecting, inserting, tightening or soldering whatever product the factory is manufacturing. While assembly is one of the oldest and widest applications of robotics, effectively simulating its many high-precision and contact-rich interactions remains a challenging task.

In the new paper Factory: Fast Contact for Robotic Assembly, a research team from NVIDIA Corporation and the University of Washington introduces Factory, a set of physics simulation methods and robot learning tools for simulating contact-rich interactions in assembly with high accuracy, efficiency, and robustness.

The team summarizes their study’s main contributions as:

1. A physics simulation module for fast, accurate simulations of contact-rich interactions through a novel synthesis of signed distance function (SDF)-based collisions, contact reduction, and a Gauss-Seidel solver.
2. A robot learning suite consisting of a Franka robot and all rigid-body assemblies from the NIST Assembly Task Board 1, the established benchmark for robotic assembly.
3. Proof-of-concept RL policies for a simulated Franka robot to solve the most contact-rich task on the NIST board, nut-and-bolt assembly.

The researchers set out to build a tool that enables fast, accurate, and robust robotic assembly simulation with three key considerations: 1) geometric representations, 2) contact reduction schemes, and 3) numerical solvers.

For geometric representation, the team adopted discrete, voxel-based SDFs to map points to distance-to-a-surface and ensure efficient, robust collision detection. For contact reduction, they combined normal similarity, penetration depth, and an area-based metric to reduce contacts and demonstrate the desired dynamics properties across various evaluation scenes. The team chose the Gauss-Seidel method as their numerical solver, as it can be accelerated via contact reduction to achieve better performance than other popular solvers such as Jacobi.

The team’s resulting suite of physics simulation methods and robot learning tools — which they dub “Factory” — is able to simulate thousands of contact-rich interactions in PhysX and Isaac Gym environments in real-time on a single GPU. The researchers also provide 60 carefully designed, ISO-standard or manufacturer-based assets from the NIST Assembly Task Board 1 for high-accuracy simulation; and train proof-of-concept reinforcement learning (RL) policies in Isaac Gym for contact-rich nut-and-bolt assembly.

Although Factory was designed to establish a state-of-the-art for contact-rich simulation in robotic assembly, the researchers say it can also be applied to additional robotics tasks such as grasping of complex non-convex shapes in home environments, locomotion on uneven outdoor terrain, and non-prehensile manipulation of aggregates of objects. The researchers hope that their work can help accelerate the efficiency of robotic assembly, and invite the machine learning community to establish benchmarks for solving the provided scenes and extend Factory to their own contact-rich applications.

A Factory demo video is available on Vimeo. The paper Factory: Fast Contact for Robotic Assembly is on arXiv.

---

Author: Hecate He | Editor: Michael Sarazen

---

We know you don’t want to miss any news or research breakthroughs. Subscribe to our popular newsletter Synced Global AI Weekly to get weekly AI updates.