Stateconditioned
State-conditioned is an adjective used to describe models, processes, or systems whose outputs or decisions depend on the current state of the subject being modeled. In fields such as control theory, reinforcement learning, and generative modeling, conditioning on state means that state variables—such as position, velocity, sensor readings, or latent state—are incorporated when producing results. The term emphasizes that behavior is dynamic and contingent on evolving state rather than fixed or state-free.
In control and reinforcement learning, a state-conditioned policy π(a|s) selects actions based on the current state
Techniques to implement state conditioning include concatenating state vectors with input observations, using conditioning mechanisms like
Challenges arise from high-dimensional or partially observed states, non-stationary environments, and the need to generalize to
See also: state-space models, conditioning (statistics), Markov decision processes, policy. The concept of state-conditioning can be