Learns high-level symbolic state machines directly from raw pixels to guide robot control without hand-crafted priors.
March 30, 2026
Original Paper
Emergent Neural Automaton Policies: Learning Symbolic Structure from Visuomotor Trajectories
arXiv · 2603.25903
The Takeaway
It bridges the gap between end-to-end black-box policies and symbolic AI. By automatically discovering discrete task structures (Mealy machines) from video, it improves sample efficiency by 27% and provides interpretable robot 'intent' maps.
From the abstract
Scaling robot learning to long-horizon tasks remains a formidable challenge. While end-to-end policies often lack the structural priors needed for effective long-term reasoning, traditional neuro-symbolic methods rely heavily on hand-crafted symbolic priors. To address the issue, we introduce ENAP (Emergent Neural Automaton Policy), a framework that allows a bi-level neuro-symbolic policy adaptively emerge from visuomotor demonstrations. Specifically, we first employ adaptive clustering and an e