MABEL: Cost-Effective Mobile Robot for Advanced Dexterity Tasks

Paper: MABEL: An Open Anthropomorphic Mobile Bimanual Platform for Teleoperation and Learning

Authors: Qilong Cheng, Rooholla Khorrambakht, Mohamed Yassine Kabouri, Quang Nam Nguyen, Luca Macesanu, Haoran Zhou, Yipeng Wang, Ludovic Righetti

Executive Summary

The research introduces MABEL, an innovative open-source mobile robot designed for tasks requiring advanced mobility and dexterity. With a total parts cost of $8,100, MABEL stands out by integrating features typically found in more expensive commercial robots. It boasts 56 degrees of freedom (DOF), consisting of two seven-DOF arms and hands with 17 DOF, as well as a pitching torso, a lifting mechanism, a three-DOF neck, and a holonomic base that allows for smooth movement in any direction.

MABEL's design prioritizes versatility and adaptability for various applications. The robot's back-drivable arms and torso facilitate compliant interactions, which may enhance its effectiveness in tasks that require gentle handling or nuanced movements. Its reach spans from the floor to a height of 2.23 meters, making it suitable for a range of environments and tasks.

A notable feature of MABEL is its enhanced perception capabilities. The robot’s head and torso configuration allows for nearly three times the camera coverage compared to traditional robots that are limited to a fixed torso camera. This increased visibility could improve the robot's ability to navigate complex environments and interact with objects more effectively.

MABEL employs advanced technology for control and operation. It utilizes Apple Vision Pro for head and hand tracking, enabling a form of teleoperation that does not require handheld controllers. This could make the robot more intuitive and easier to operate for users, improving usability in practical applications. The system achieves a mean latency of 137 milliseconds for video transmission, ensuring responsive control during operation.

The research also details an end-to-end imitation learning pipeline. This process starts with collecting data via teleoperation and concludes with deploying learned policies through a shared command interface and whole-body controller. Such a pipeline could streamline the development of robotic tasks, allowing for quicker adaptation to new environments or objectives.

While the study primarily presents findings from simulations and benchmarks, the implications for real-world applications are significant. MABEL could be adapted for use in various sectors, including manufacturing, logistics, and healthcare, where tasks often require fine motor skills and the ability to navigate dynamic workspaces. The open-source nature of MABEL may also encourage further innovation and customization by researchers and developers in the robotics community.

Overall, MABEL represents a promising advancement in the field of robotics, particularly in making advanced robotic capabilities more accessible and cost-effective for diverse applications.

Academic Abstract

As robot learning moves beyond the tabletop, many tasks require mobility, adjustable reach, active perception, and bimanual dexterity. Commercial platforms combining these capabilities are costly, while affordable open-source alternatives cover only a subset and require software integration. We present MABEL, an open-source anthropomorphic mobile bimanual robot with 56 degrees of freedom (DOF) and a US$8.1k parts cost. It combines two 7-DOF arms with 17-DOF hands, a pitching torso, a lift, a 3-DOF neck, and a holonomic base. Back-drivable arms and torso support compliant interaction, with fingertip reach from floor level to 2.23 m. The head and torso roughly triple modeled camera coverage relative to a torso-fixed camera. Apple Vision Pro head and hand tracking enables whole-body teleoperation without handheld controllers, with 137 ms mean glass-to-glass stereo video latency over the robot's Wi-Fi hotspot. We demonstrate an end-to-end imitation learning pipeline, from teleoperated data collection to policy deployment, using a shared command interface and whole-body controller.