Use of forward models for control versus planning in model-based reinforcement learning

Determine whether output-input (forward) models used in model-based reinforcement learning for robotic control are employed for ongoing motor control and state estimation, rather than solely for planning, and identify empirical criteria to distinguish these roles.

Background

The embodiment indicator emphasizes using output-input models for perception and motor control (e.g., state estimation, feedback correction), distinguishing endogenous from exogenous changes. While forward models are common in model-based RL for robotics, their specific use as control mechanisms versus planning tools remains uncertain.

Clarifying this will inform whether current or near-term embodied agents meet the embodiment criteria relevant to consciousness indicators.