BEHAVIOR Robot Suite: Streamlining Real-World Whole-Body Manipulation for Everyday Household Activities (2503.05652v1)

Abstract: Real-world household tasks present significant challenges for mobile manipulation robots. An analysis of existing robotics benchmarks reveals that successful task performance hinges on three key whole-body control capabilities: bimanual coordination, stable and precise navigation, and extensive end-effector reachability. Achieving these capabilities requires careful hardware design, but the resulting system complexity further complicates visuomotor policy learning. To address these challenges, we introduce the BEHAVIOR Robot Suite (BRS), a comprehensive framework for whole-body manipulation in diverse household tasks. Built on a bimanual, wheeled robot with a 4-DoF torso, BRS integrates a cost-effective whole-body teleoperation interface for data collection and a novel algorithm for learning whole-body visuomotor policies. We evaluate BRS on five challenging household tasks that not only emphasize the three core capabilities but also introduce additional complexities, such as long-range navigation, interaction with articulated and deformable objects, and manipulation in confined spaces. We believe that BRS's integrated robotic embodiment, data collection interface, and learning framework mark a significant step toward enabling real-world whole-body manipulation for everyday household tasks. BRS is open-sourced at https://behavior-robot-suite.github.io/

BEHAVIOR Robot Suite: Enabling Autonomous Whole-Body Manipulation for Household Tasks

The paper "BEHAVIOR Robot Suite: Streamlining Real-World Whole-Body Manipulation for Everyday Household Activities" addresses the challenge of enabling mobile robots to autonomously perform whole-body manipulation tasks in everyday household environments. This work is grounded in the analysis of the BEHAVIOR-1K benchmark, which catalogs 1,000 human-centered household activities to isolate critical capabilities for successful task completion. These capabilities include bimanual coordination, stable and precise navigation, and extensive end-effector reachability.

Framework Overview

The BEHAVIOR Robot Suite (BRS) is proposed as an integrated framework to learn and execute whole-body manipulation policies that leverage these capabilities. BRS comprises two key innovations:

1. JoyLo Interface: A low-cost, whole-body teleoperation system that facilitates data collection crucial for visuomotor policy development. Designed specifically for the Galaxea R1 robot, JoyLo combines 3D-printed leader arms with Nintendo Joy-Con controllers offering rich feedback and precise control. This mechanism allows seamless teleoperation, collecting high-quality, singularity-free data paramount for imitation learning methods.
2. Whole-Body VisuoMotor Attention (WB-VIMA) Policy: A novel learning algorithm that models coordinated whole-body actions by leveraging the robot's hierarchical embodiment structure. WB-VIMA employs autoregressive action denoising and multi-modal observation attention, mitigating the challenges associated with modeling complex whole-body actions in high-dimensional spaces.

Empirical Evaluation

The BRS framework is evaluated on five representative household tasks, exhibiting its capability to autonomously complete challenging multi-stage activities. The success rates across tasks demonstrate the system's ability to generalize in unmodified human environments, achieving average success rates of 58% and peak rates of 93%. These results surpass human teleoperation on tasks demanding precise control of contact interactions, underscoring the efficacy of the hierarchical approach in WB-VIMA.

Moreover, quantitative comparisons reveal JoyLo's superiority over VR controllers and Apple Vision Pro in data collection efficiency and task completion rates. JoyLo's physical embodiment constraints prevent infeasible actions, significantly increasing the replay success rate—a crucial metric for reliable policy training.

Implications and Future Directions

The successful integration of advanced teleoperation interface technology and hierarchical action modeling in the BRS framework signifies progression towards autonomous robotic systems capable of complex household tasks. Practically, BRS can be adopted for refined manipulation in diverse unstructured environments, advancing domains such as assistive technology where adaptive and reliable robot behavior is essential.

Theoretically, this work invites exploration into scalability and embodiment-transfer capabilities, questioning how techniques like multi-robot training data and modalities like synthetic and human-provided datasets could further enhance robot autonomy and scene-level generalization.

As AI continues evolving, the methodologies proposed in this paper chart pathways in robotics research focused on engaging complex real-world environments, making substantial contributions to whole-body manipulation capabilities and their practical applications.