Construction of Machine Self-Awareness in Embodied Intelligence
Machine self-awareness remains a central but un-resolved challenge in embodied intelligence. Existing artificial intelligence systems are largely built on an outside-in paradigm, in which external sensory data are processed into task-oriented outputs. Although such systems can support perception, prediction, and control, they lack an internally grounded mechanism for distinguishing self-generated change from environmental disturbance. This paper proposes an inside-out developmental framework for constructing machine self-awareness. Drawing on embodied cognition, the C0–C2 hierarchy of consciousness, the efference copy mechanism, and the Free Energy Principle, the framework argues that self-awareness should emerge through repeated cycles of motor intention, predictive modeling, sensory feedback, and prediction-error correction. At the core of this process is the formation of a stable self/environment boundary, defined not by anatomical limits but by the coherence of predictive control. The paper develops an eight-stage model that begins with un-conscious information processing, progresses through efference-copy-based sensorimotor calibration and boundary formation, and extends toward global integration, spatial cognition, active perception, and metacognitive self-monitoring. To illustrate this developmental logic, the paper introduces The Chrysalis of Self, a 2D interactive prototype that simulates mirror-test recognition and spatial memory exploration. The prototype is presented as a pedagogical demonstration rather than an empirical validation. The paper further discusses architectural implications, including the limitations of digital twins, the possibility of tool-based extended selfhood, and the need for multi-criteria evaluation combining structural, boundary, and metacognitive evidence. Overall, this work argues that machine self-awareness cannot be externally assigned; it must be progressively constructed through embodied action, prediction, and feedback.