English

$\pi2\text{vec}$: Policy Representations with Successor Features

Machine Learning 2024-01-25 v2 Robotics

Abstract

This paper describes π2vec\pi2\text{vec}, a method for representing behaviors of black box policies as feature vectors. The policy representations capture how the statistics of foundation model features change in response to the policy behavior in a task agnostic way, and can be trained from offline data, allowing them to be used in offline policy selection. This work provides a key piece of a recipe for fusing together three modern lines of research: Offline policy evaluation as a counterpart to offline RL, foundation models as generic and powerful state representations, and efficient policy selection in resource constrained environments.

Cite

@article{arxiv.2306.09800,
  title  = {$\pi2\text{vec}$: Policy Representations with Successor Features},
  author = {Gianluca Scarpellini and Ksenia Konyushkova and Claudio Fantacci and Tom Le Paine and Yutian Chen and Misha Denil},
  journal= {arXiv preprint arXiv:2306.09800},
  year   = {2024}
}

Comments

Accepted paper at ICLR2024

R2 v1 2026-06-28T11:07:08.981Z