Intrinsically motivated agents replicate fruit fly exploration, predicting actions 10s ahead

NeuroAgentsLab · hf · 2026-10-27

A University of Maryland / CMU team posted a bioRxiv paper using hierarchical intrinsically motivated agents to reverse-engineer fruit fly autonomy:

Whole-brain recordings from Drosophila melanogaster identify neural representations of spontaneous exploratory decisions more than 10 seconds in advance;
Combined with a neuromechanical digital twin, the team trains hierarchical agents with competency-based intrinsic objectives and realistic embodiments;
These agents quantitatively match fly exploratory behavior and internal representations of future volitional action;
The work supports optimizing intrinsic goals as a route to engineering autonomous decision-making, with a biological ground truth as benchmark.

More from AGI Musings