Event
Authors propose ENAP neuro-symbolic imitation learning framework
1 reports · 1 sources
Yiyuan Pan and other authors propose the ENAP (Emergent Neural Automaton Policy) neuro-symbolic imitation learning framework: using adaptive clustering and an extension of the L* algorithm, it learns a Mealy state machine from demonstrations as an interpretable high-level planner, and this discrete structure then guides a low-level reactive residual network, learning continuous control through behavioral cloning. The authors claim that, under low-data conditions, ENAP outperforms current end-to-end vision-language-action (VLA) policies by up to 27%; ENAP can also extract task structure from the latent space of a pretrained VLA, surpassing that VLA on the CALVIN benchmark with 39% fewer parameters. These comparison results are all as stated by the authors.
Report timeline
- ENAP: Neural Automaton Policies That Learn Symbolic Structure from Visuomotor Trajectories
The authors propose ENAP (Emergent Neural Automaton Policy), a neuro-symbolic imitation learning framework that uses adaptive clustering and an extension of the L* algorithm to learn a Mealy state machine from demonstrations as an interpretable high-level planner, which then guides a low-level reactive residual network to learn continuous control via behavior cloning.
arXiv cs.RO ·