Why does ArticulationBody drift over time?

The solver applies position correction each step but with limited iteration count, residual errors accumulate. Higher iteration counts reduce drift at perf cost. Articulation drives are stiffer than legacy ConfigurableJoint, but solver iteration still matters.

How many iterations are enough?

Default 6 is fine for simple chains. For robotic arms or complex mechanisms, 20–30 is common. Above 50, perf cost becomes meaningful. Profile in the Physics Debugger to find the right balance.

Should I use ArticulationBody or Rigidbody+ConfigurableJoint?

ArticulationBody for kinematic chains (robotics, vehicles, ragdolls with predictable behavior). Rigidbody+ConfigurableJoint for one-off constraints with simpler setup. Articulation is more stable; Configurable is more flexible.

Fix: Unity ArticulationBody Fixed Joint Drift

Quick answer: Set ArticulationBody.solverIterations to 20+ on each body. Lower fixedDeltaTime to 0.01 if needed. Articulation drift is a solver iteration count problem.

A robotic arm built with ArticulationBody has its end effector slowly drift away from the intended pose over a minute of simulation. Small per-tick errors compound. The arm is built correctly; the solver isn’t converging tight enough.

Articulation Solver

ArticulationBody uses PhysX’s articulated joint solver, which is a reduced-coordinate formulation tighter than ConfigurableJoint but still iterative. Each tick, the solver runs N iterations refining positions; with too few iterations, residual constraint error remains and accumulates as drift.

The Fix

// On each body in the chain
articulationBody.solverIterations = 24;
articulationBody.solverVelocityIterations = 8;

Default solverIterations is 6, velocity is 1. Bumping to 24/8 dramatically reduces drift for typical robotic chains. Cost: roughly 3× articulation CPU per body.

Project-Level Defaults

Edit → Project Settings → Physics → Default Solver Iterations. Setting these higher applies to all new ArticulationBodies. Per-body overrides as above for specific tight chains.

Fixed Timestep

Lower fixedDeltaTime from 0.02 to 0.01:

Time.fixedDeltaTime = 0.01f;

Doubles simulation steps per second; halves per-step error. Cost: double physics CPU. Combine with iteration tuning for stable robotic motion.

Joint Drives

For positional drives, the Force Limit and Damping play a role. Lower damping = bouncier; higher = slower but stable. For robotics, start with Damping = Force/10.

Verifying

Set a target pose. Let the simulation run for 60 seconds. Drift should be sub-millimeter. With default settings, drift is often visible (centimeters). Compare runs with and without tuning to confirm.

“Articulation solver iterations are the main lever for drift. Default 6 is too few for tight chains; 20+ is normal for robotics.”

For VR hands or precise IK setups, even higher iteration counts (32–48) are common — players notice millimeter drift on their own hands.