Principle:Isaac sim IsaacGymEnvs Asset Preparation
| Knowledge Sources | |
|---|---|
| Domains | Simulation, Manipulation |
| Last Updated | 2026-02-15 00:00 GMT |
Overview
Process of loading URDF meshes and computing Signed Distance Field representations for contact-rich robotic assembly simulation.
Description
Contact-rich assembly tasks (nut-bolt, plug-socket, gears) require accurate collision geometry beyond what standard physics engines provide. IsaacGymEnvs loads URDF/OBJ mesh files, samples surface points, and creates Warp mesh objects for SDF queries. These SDFs enable reward shaping based on penetration depth and engagement detection.
The asset preparation pipeline consists of three stages:
- Mesh Loading: URDF files are parsed to extract visual and collision meshes. OBJ mesh files referenced in the URDF are loaded using trimesh, providing vertex positions and face indices.
- Surface Point Sampling: Points are uniformly sampled on the mesh surface. These sampled points serve as query locations for SDF evaluation during reward computation.
- Warp Mesh Construction: The loaded mesh vertices and faces are used to construct NVIDIA Warp mesh objects. Warp provides GPU-accelerated SDF queries via
wp.mesh_create()andwp.mesh_query_point().
Usage
Use this principle when setting up Factory/IndustReal assembly environments that need fine-grained contact information. Specifically:
- IndustReal plug-socket insertion: SDF queries determine how deeply the plug has entered the socket, enabling smooth reward gradients for insertion tasks.
- IndustReal gear meshing: SDF-based rewards guide gear teeth into proper engagement.
- Factory nut-bolt threading: Surface point sampling enables detection of nut-to-bolt engagement.
Asset preparation is a one-time setup cost performed during environment initialization, not during the training loop.
Theoretical Basis
Signed Distance Fields provide a continuous representation of proximity to a surface:
# Signed Distance Field definition
For a point p and surface S:
SDF(p) = +d if p is outside S (distance d to nearest surface point)
SDF(p) = 0 if p is on S
SDF(p) = -d if p is inside S (penetration depth d)
SDF queries enable three critical capabilities for assembly reward shaping:
- Penetration detection:
SDF(p) < 0indicates the point is inside the mesh, with magnitude indicating depth. - Surface proximity rewards: Smooth reward gradients as objects approach contact, avoiding sparse reward problems.
- Engagement verification: By querying SDF of plug surface points against the socket mesh, the fraction of points with
SDF < 0indicates insertion progress.
The mathematical formulation for SDF-based reward:
R_sdf = (1/N) * sum_{i=1}^{N} SDF_socket(T(p_plug_i))
where:
N = number of sampled surface points on the plug
T(p) = rigid body transform of point p from plug frame to socket frame
SDF_socket = signed distance field of the socket mesh