Quantization-Free Autoregressive Action Transformer

TL;DR

Current transformer-based imitation learning approaches introduce discrete action representations and train an autoregressive transformer decoder on the resulting latent code…

Opening excerpt from the authors’ abstract. source