OpenHEART: Opening Heterogeneous Articulated Objects with a Legged Manipulator
Seonghyeon Lim, Hyeonwoo Lee, Seunghyun Lee, I Made Aswin Nahrendra, Hyun Myung
AI summary
Problem
Legged manipulators struggle to robustly interact with heterogeneous articulated objects due to diverse joint types and complex robot dynamics. Existing reinforcement learning methods rely on high-dimensional sensory inputs, leading to poor sample efficiency and limited generalization.
Approach
The authors propose a hierarchical RL framework that compresses object geometry into low-dimensional features via SAFE and adaptively fuses visual and proprioceptive data with ArtIEst to estimate opening parameters.
Key results
- First autonomous framework to open heterogeneous articulated objects with a legged manipulator without precise object models
- ArtIEst reduces articulation estimation error by adaptively mixing exteroception and proprioception
- SAFE mitigates overfitting and improves cross-domain generalization through low-dimensional geometric encoding
- Successfully validated in simulation and real-world experiments across diverse object types
Why it matters
Enables legged robots to perform versatile, real-world manipulation tasks in unstructured environments, advancing autonomous mobile manipulation.
Abstract
Legged manipulators offer high mobility and ver- satile manipulation. However, robust interaction with heteroge- neous articulated objects, such as doors, drawers, and cabinets, remains challenging because of the diverse articulation types of the objects and the complex dynamics of the legged robot. Existing reinforcement learning (RL)-based approaches often rely on high-dimensional sensory inputs, leading to sample inef- ficiency. In this paper, we propose a robust and sample-efficient framework for opening heterogeneous articulated objects with a legged manipulator. In particular, we propose Sampling-based Abstracted Feature Extraction (SAFE), which encodes handle and panel geometry into a compact low-dimensional repre- sentation, improving cross-domain generalization. Additionally, Articulation Information Estimator (ArtIEst) is introduced to adaptively mix proprioception with exteroception to estimate opening direction and range of motion for each object. The proposed framework was deployed to manipulate various het- erogeneous articulated objects in simulation and real-world robot systems. Videos can be found on the project website: https://openheart-icra.github.io/OpenHEART/.