Learning Awareness Models


Nov 07, 2017 (modified: Nov 07, 2017) ICLR 2018 Conference Blind Submission readers: everyone Show Bibtex
  • Abstract: We consider the setting of an agent with a fixed body interacting with an unknown and uncertain external world. In this setting we show how predictive models of proprioception can be used to reason about interactions between our agent and external objects. Such models are appealing because they can be fit using data that is readily available to the agent through the course of its existence without requiring access to privileged information about the state of the external world. In spite being trained with only internally available signals these predictive models come to represent external objects through the necessity of predicting their effects on the agent's own body. We demonstrate this in simulation by using the proprioceptive models to make predictions about properties of external objects. We also collect data from a real robotic platform and show that the same models can be used to answer questions about properties of objects in the real world.
  • TL;DR: We train predictive models on proprioceptive information and show they represent properties of external objects.
  • Keywords: Awareness, Prediction, Seq2seq, Robots