Humanoid robots learn to carry objects safely in cluttered spaces
HOTICE: Whole-Body Humanoid Object Transportation in Cluttered Environments
Robotics
Summary
Carrying objects through tight, cluttered spaces is hard for robots because they need to avoid bumping into things while holding items. The authors created HOTICE, a new system that helps humanoid robots move both their bodies and the items they carry carefully around obstacles. They made robots smarter by splitting control between upper and lower body parts and teaching them with examples of expert behaviors. Their approach works well in simulations and on a real robot, handling different shapes and layouts.
What this means in practice
- •For robotic delivery teams: Enable humanoid delivery robots to carry packages reliably through crowded indoor environments without dropping or bumping objects.$Commercial implications: This technology allows companies selling delivery robots to offer robust object handling in cluttered real-world spaces, improving service reliability.
- •For warehouse automation engineers: Design warehouse robots that can transport various items safely through narrow aisles and crowded workspaces by coordinating whole-body motions.
Authors
Toan Nguyen, Weiduo Yuan, Siheng Zhao, Yue Wang, Daniel Seita
Abstract
Object transportation is a fundamental capability for humanoid robots operating in real-world, human-centric environments, yet existing methods struggle when clutter constrains free space around both the robot and its carried payload. We present HOTICE, a whole-body humanoid learning framework for transporting objects through such cluttered environments. First, we introduce Humanoid-Object Decoupled Potential Fields, which jointly encode collision-avoidance guidance for the robot and the carried object, enabling coordinated, obstacle-aware motion for both. Second, to address the large action space inherent to whole-body loco-manipulation, we design a dual-agent reinforcement learning architecture that decouples upper- and lower-body control while preserving whole-body coordination via shared state observations and rewards. To train a policy that generalizes across diverse cluttered scenes, we further employ a specialist-to-generalist distillation strategy, in which privileged teacher policies are distilled into a single deployable student policy. We evaluate HOTICE in MuJoCo simulation and on a real Unitree G1 humanoid, demonstrating effective and robust object transportation across cluttered scenarios for objects of varying shapes. Our results show that HOTICE reliably coordinates whole-body motion and object-aware collision avoidance, generalizing effectively to previously unseen cluttered environments while achieving strong performance in sim2real deployment.