Peter Stone's Selected Publications

Classified by TopicClassified by Publication TypeSorted by DateSorted by First Author Last NameClassified by Funding Source


SLAC: Simulation-Pretrained Latent Action Space for Whole-Body Real-World RL

SLAC: Simulation-Pretrained Latent Action Space for Whole-Body Real-World RL.
Jiaheng Hu, Peter Stone, and Roberto Martín-Martín.
In Conference on Robot Learning (CoRL), September 2025.

Download

[PDF]2.6MB  

Abstract

Building capable household and industrial robots requires mastering the controlof versatile, high-degree-of-freedom (DoF) systems such as mobile manipulators.While reinforcement learning (RL) holds promise for autonomously acquiring robotcontrol policies, scaling it to high-DoF embodiments remains challenging. DirectRL in the real world demands both safe exploration and high sample efficiency,which are difficult to achieve in practice. Sim-to-real RL, on the other hand, isoften brittle due to the reality gap. This paper introduces SLAC, a method thatrenders real-world RL feasible for complex embodiments by leveraging alow-fidelity simulator to pretrain a task-agnostic latent action space. SLACtrains this latent action space via a customized unsupervised skill discoverymethod designed to promote temporal abstraction, disentanglement, and safety,thereby facilitating efficient downstream learning. Once a latent action space islearned, SLAC uses it as the action interface for a novel off-policy RL algorithmto autonomously learn downstream tasks through real-world interactions. Weevaluate SLAC against existing methods on a suite of bimanual mobile manipulationtasks, where it achieves state-of-the-art performance. Notably, SLAC learnscontact-rich whole-body tasks in under an hour of real-world interactions,without relying on any demonstrations or hand-crafted behavior priors.

BibTeX Entry

@InProceedings{jiaheng_hu_2025,
  author   = {Jiaheng Hu and Peter Stone and Roberto Martín-Martín},
  title    = "{SLAC}: Simulation-Pretrained Latent Action Space for Whole-Body Real-World {RL}",
  booktitle = {Conference on Robot Learning (CoRL)},
  year     = {2025},
  month    = {September},
  location = {Seoul, Korea},
  abstract = {Building capable household and industrial robots requires mastering the control
of versatile, high-degree-of-freedom (DoF) systems such as mobile manipulators.
While reinforcement learning (RL) holds promise for autonomously acquiring robot
control policies, scaling it to high-DoF embodiments remains challenging. Direct
RL in the real world demands both safe exploration and high sample efficiency,
which are difficult to achieve in practice. Sim-to-real RL, on the other hand, is
often brittle due to the reality gap. This paper introduces SLAC, a method that
renders real-world RL feasible for complex embodiments by leveraging a
low-fidelity simulator to pretrain a task-agnostic latent action space. SLAC
trains this latent action space via a customized unsupervised skill discovery
method designed to promote temporal abstraction, disentanglement, and safety,
thereby facilitating efficient downstream learning. Once a latent action space is
learned, SLAC uses it as the action interface for a novel off-policy RL algorithm
to autonomously learn downstream tasks through real-world interactions. We
evaluate SLAC against existing methods on a suite of bimanual mobile manipulation
tasks, where it achieves state-of-the-art performance. Notably, SLAC learns
contact-rich whole-body tasks in under an hour of real-world interactions,
without relying on any demonstrations or hand-crafted behavior priors.
  },
}

Generated by bib2html.pl (written by Patrick Riley ) on Sat Oct 04, 2025 14:51:47