Principle:Google deepmind Mujoco Collision Detection

Knowledge Sources	Domains	Last Updated
Google DeepMind MuJoCo	Physics Simulation, Geometry	2025-02-15

Overview

Description: MuJoCo's collision detection system identifies contacts between geometric bodies in the simulation. The pipeline includes broadphase filtering (bounding volume hierarchy), midphase pair reduction, and narrowphase geometry-specific algorithms.

Context: Collision detection is a critical component of physics simulation. MuJoCo implements multiple narrowphase algorithms: closed-form solutions for primitive pairs (sphere, capsule, box, cylinder), GJK/EPA for general convex pairs, SDF-based methods for mesh collisions, and specialized box-box intersection.

Theoretical Basis

Collision detection in rigid body simulation typically follows a three-phase approach:

• Broadphase: Axis-aligned bounding box (AABB) overlap tests to quickly cull non-interacting pairs
• Midphase: Bounding volume hierarchy (BVH) traversal for mesh geometries
• Narrowphase: Exact contact computation using geometry-specific algorithms (GJK, EPA, SAT, SDF gradient descent)

MuJoCo supports geom types including plane, hfield, sphere, capsule, ellipsoid, cylinder, box, and mesh. Each pair type may use a different narrowphase algorithm.

Related Pages

Implementations

• Implementation:Google_deepmind_Mujoco_Engine_Collision_Box
• Implementation:Google_deepmind_Mujoco_Engine_Collision_Convex
• Implementation:Google_deepmind_Mujoco_Engine_Collision_Driver
• Implementation:Google_deepmind_Mujoco_Engine_Collision_GJK
• Implementation:Google_deepmind_Mujoco_Engine_Collision_Primitive
• Implementation:Google_deepmind_Mujoco_Engine_Collision_SDF

Workflows

• (none yet)