HANDLE: Robust Non-prehensile Liquid Manipulation in Cluttered, Human-centric Environments with Physical HRI

Yuhe Gong, junhui huang, Riddhiman Laha, Zewen Yang, Hao Xing, Ayse Kucukyilmaz, Seth Hutchinson, Luis Figueredo

Keywords: Robust Robot Manipulation, Human–Robot Interaction, Physical Human Guidance, Disturbance Rejection, Constraint-Based Control, Whole-Body Reactive Control, Slosh Free Transport

TL;DR: HANDLE is a control framework that allows robots to safely manipulate under highly robust and non-prehensive tasks, even while navigating cluttered enviornments, reacting to physical human contact, and following unpredictable movement commands.

Abstract:

Robust manipulation in human-centered environ-
ments requires handling tightly coupled sources of uncertainty,
including dynamic obstacles, physical human interaction, and
payload-induced constraints. This challenge is particularly acute
in non-prehensile liquid transport, where external disturbances
directly affect both task feasibility and system stability. We present
HANDLE (Human-Aware Non-prehensile hanDling of Liquids
with whole-body rEactive Control), a real-time framework for
robust manipulation under coupled dynamic and interaction
disturbances. HANDLE employs a hierarchical robustness archi-
tecture that explicitly addresses three complementary aspects of
robustness: (i) environmental robustness via a geometric-aware
safety layer enabling real-time whole-body collision avoidance in
dynamic and cluttered scenes; (ii) dynamic robustness via a slosh-
aware control layer that regulates accelerations and orientation
to maintain fluid stability; and (iii) interaction robustness via a
compliance mechanism that accommodates direct human-applied
forces without compromising previous conditions. Central to
our approach is a constraint-consistent projection mechanism
that maps arbitrary external inputs—including teleoperation
commands and stochastic human disturbances—onto a stability-
constrained safe manifold, ensuring feasibility without sacrificing
responsiveness. We evaluate the proposed framework under
aggressive and unpredictable conditions, including abrupt velocity
commands, strong physical human guidance, and simultane-
ous multi-modal physical contacts and disturbances in shared
workspaces. Results demonstrate that HANDLE maintains con-
straint satisfaction, prevents spillage and collisions, and preserves
stable non-prehensile transport under conditions where state-of-
the-art methods fail. A user study further indicates improved
perceived safety, robustness, and operational intuitiveness.

Submission Number: 43