Initial state and reset semantics
A repeated command sequence is not a repeated experiment unless the starting state is controlled.
Research note · November 14, 2026Research question
Which state must be restored before comparing two simulation runs?
Method
- List positions, velocities, controller memory, actuator state and other persistent values for a bounded model. Separate model parameters from dynamic state.
- Run a command sequence twice with a full reset, then twice with only positions reset. Compare trajectories and identify which retained state explains any difference.
- Preserve the reset procedure with the run record. Treat a reset failure or unsupported hidden state as an experimental limitation rather than assuming a fresh start.
Records to retain
- Initial positions and velocities
- Controller and actuator state
- Reset procedure revision
- Trajectory and reset diagnostics
Source basis
MuJoCo’s simulation documentation distinguishes integration state, simulation state and state-manipulation operations. Reproducibility requires more than returning the visible geometry to its starting pose.
Interpretation & limits
Third-party components may have hidden state that cannot be serialized. This protocol proposes a controlled comparison; it does not claim arbitrary connected devices can be reset reproducibly.
Next experiments
- Build a fixture with a controller integrator that survives a partial reset.
- Define which state is included in a saved experiment.
References
Technical reference
Implementation details supporting this note, drawn directly from the maintained documentation. These are technical references, not additional papers or new experimental results. Expand a topic to inspect its method, fields and limits.