finding

active

finding:agent-achieves-approximately-70-sticker-removal-success-rate-by-end-of-500k-training-steps

Agent achieves approximately 70% sticker-removal success rate by end of 500k training steps

Main behavioral result demonstrating the model's efficacy in the mirror-mark task

Source paper

extracted_from

Active Inference with a Self-Prior in the Mirror-Mark Task

(2026) · Dongmin Kim · Hoshinori Kanazawa · Yasuo Kuniyoshi

Neighborhood — ranked by edge-count

Claims (3)

claim

A sensory anomaly (sticker mismatch) can itself become an intrinsic drive for action under active inference, even without external reward
associated_withsupports
Core mechanism claim linking mismatch detection to behavior through EFE minimization
The self-prior operates as an internal criterion for distinguishing self from non-self, without external reward or explicitly computed sticker location
supports
Central interpretive claim of the paper, supported by EFE decrease after sticker removal
Failure episodes are primarily caused by hand occluding the sticker in the mirror or sticker leaving the visual field due to head rotation, plus kinematic reachability limits
cites
Explains the ceiling on removal success as due to perceptual and kinematic constraints, not principled failures

Questions (1)

question

How can mark-directed behavior emerge from the agent's own experience alone, without external reward or an explicit goal?
answered_by
The central research question motivating the paper

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

Sticker-removal success rate stayed near 20% in the early phase of trainingfinding0.892
Shows learning progression from chance-level to functional behavior
Mean hand-sticker distance decreased gradually across 500k training steps, including before removal probability exceeded 50%finding0.817
Suggests the agent learned to recognize and approach the sticker before achieving reliable removal
Sticker Removal Success Criterionmethod0.795
Operational definition: hand stays within 2 cm of sticker for 50 consecutive steps (0.5 seconds)
Untrained model (0 training steps) shows no clear EFE difference before and after sticker removal (Δ = +1.70)finding0.780
Control showing that the EFE signal is learned, not inherent to the architecture
In a single illustrative episode (seed 2), mean EFE after sticker removal was 12.00 lower than before removal after 500k training stepsfinding0.769
Qualitative confirmation of EFE drop in trained model vs. untrained model (Δ = +1.70)
The sticker-removal behavior induced by the self-prior corresponds to stimulus-elicited intention rather than endogenous intention, aligning with the developmental view of early intentional agencyclaim0.743
Connects the model's behavior to Zaadnoordijk and Bayne's taxonomy of intentional agency
EFE decrease after sticker removal is statistically significant (Wilcoxon p = 6.33×10⁻⁹) across 80 evaluationsfinding0.740
Confirms that EFE systematically decreases after sticker removal, validating the self-prior as internal criterion
Mean EFE before sticker removal across 80 evaluations: 79.33 ± 4.34finding0.737
Baseline EFE when sticker is present, used for comparison