A novel architecture is proposed that addresses sample efficiency in VLAs by training a predictive world model on the embedding space of the VLA's vision encoder, and it is hypothesized that these embeddings are action-relevant and usable for future prediction.
Parsa Mastouri Kashani, Jan-Gerrit Habekost, Stefan Wermter· 0 citations
SmoLSTM couples a frozen 256M-parameter SmolVLM backbone to a matrix-memory LSTM control layer in which observation tokens and action queries are unified in a single causal stream that is never reset throughout the entire episode, capturing the episode in a recurrent state.
This work proposes LAVLA, a framework for latent cluster analysis of VLA models, and conducts a layer-wise study of the state-of-the-art GR00T N1.5 model, with particular focus on its action decoder.
Theodor Wulff, Sergio Lanza, Tamara Bíla et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.