A self-prior in active inference lets an agent spontaneously reach toward a tactile stimulus

The paper proposes a novel density model of an agent's own multimodal sensory experience, the "self-prior," which generates behavioral references within an active inference framework. Whereas much prior work on intrinsic motivation focuses on how exploration contributes to acquiring external rewards, here behavioral references come from an intrinsic process minimizing mismatches between average past sensory experiences and current observations, without an external reward criterion. Evidence comes from the method design and simulated-environment experiment described in the abstract; the loaded text is the abstract, with no model equations, parameters, or quantitative metrics.

In a simulated environment, the agent spontaneously reaches toward a tactile stimulus, exhibiting goal-directed behavior. The result attributes the emergence of goal-directed behavior to intrinsic motivation shaped by the agent's own sensory experience rather than to externally rewarded exploration. Evidence is an observation in a simulated environment, stated in the abstract as confirming that "the agent spontaneously reaches toward a tactile stimulus"; no sample size, control conditions, or statistics are reported.

The authors draw an analogy between this mechanism and the acquisition and utilization of a body schema through continuous interaction with the environment. The analogy places body-schema formation in early development and goal-directed action within a single free-energy-principle framework, linking developmental science and active inference modeling. This is a conceptual analogy and theoretical integration; the abstract reports no separate experimental test of body schema.

The paper argues that the study implements intrinsically motivated behavior shaped by the agent's own sensory experiences, demonstrating the spontaneous emergence of intentional behavior in early development. Relative to intrinsic motivation research centered on obtaining external rewards, the emphasis here is that intentional behavior can emerge spontaneously without external reward. Evidence is the authors' summary claim based on the simulated experiment; the abstract offers no cross-task, cross-environment, or biological-data comparison.

The work targets an agent in a simulated environment and tests whether the "self-prior" can induce goal-directed behavior without external reward; its setting corresponds to modeling body-schema formation and reaching in early development. For readers, it offers a way to connect the free energy principle, intrinsic motivation, and body schema, relevant to those studying intentional behavior in development or designing reward-free autonomous exploration mechanisms.

Because the reading scope is incomplete, with only the abstract and reference fragments available, the concrete form of the model, the simulated environment setup, evaluation metrics, and control conditions are not presented, so how broadly and under how many conditions the result holds cannot be judged. Readers may watch how the self-prior is estimated from multimodal experience, how mismatch minimization is translated into specific actions, and whether the mechanism holds in more complex environments or against biological developmental data.