TidyBot++: An Open-Source Holonomic Mobile Manipulator for Robot Learning

Abstract

Exploiting the promise of recent advances in imitation learning for mobile manipulation will require the collection of large numbers of human-guided demonstrations. This paper proposes an open-source design for an inexpensive, robust, and flexible mobile manipulator that can support arbitrary arms, enabling a wide range of real-world household mobile manipulation tasks. Crucially, our design uses powered casters to enable the mobile base to be fully holonomic, able to control all planar degrees of freedom independently and simultaneously. This feature makes the base more maneuverable and simplifies many mobile manipulation tasks, eliminating the kinematic constraints that create complex and time-consuming motions in nonholonomic bases. We equip our robot with an intuitive mobile phone teleoperation interface to enable easy data acquisition for imitation learning. In our experiments, we use this interface to collect data and show that the resulting learned policies can successfully perform a variety of common household mobile manipulation tasks.

Teleoperated Tasks

Through teleoperation, TidyBot++ can perform a wide range of common household tasks in a real home (4x speed)

Unload oven

Set the table

Tidy up kitchen

Fill watering can

Tidy up foyer

Tidy up bedroom

Enter bathroom

Tidy up countertop

Wipe countertop

Scrub bathtub

Phone Teleoperation

We open source our intuitive phone teleoperation interface, shown in action below (4x speed)

Holonomic vs. Differential Drive

Using a holonomic mobile base simplifies teleoperation and policy learning. See comparison (4x speed) below.

Differential drive base

Holonomic base

Hardware Specifications

Swappable Arm

We provide several highly-customizable reference robot designs which can be easily modified to support different arms.

Kinova

Franka

ARX5

xArm

UR5

ViperX

Payload

Though designed to carry just one arm, we found that our mobile base can also handle much higher payloads with ease.

Weight plates totaling ~270 lb (120 kg)

Five arms totaling ~200 lb (90 kg)

Passenger weighing ~150 lb (70 kg)

Traversability

Our mobile base works on a variety floor surfaces ranging from hard floor to high-pile carpet, and can traverse many common floor obstacles. Though intended for indoor use, we found that it can handle some outdoor floor obstacles as well.

Door thresholds

Elevator gaps

Speed bumps

Steel construction plates

Asphalt curb ramps

Loading ramps (6.5° inclination)

Odometry

We evaluated the odometry accuracy by driving our mobile base along a random trajectory and comparing the odometry data (dashed blue line) with motion capture (black line). The results show translation drift of less than 1 cm per meter of distance traveled, and rotation drift of less than 1° per 360° of rotation.

Repeatability

One advantage of a holonomic mobile base is that it can be directly controlled in position mode, leading to high repeatability. To test this, we recorded the demonstration shown below (4x speed) for the push chairs task and replayed it 10 times. We found that simple open-loop replay from a fixed initial configuration achieves 10/10 success for this task.

Video

Poster

Acknowledgments

We would like to thank Kevin Zakka, Yixuan Huang, Kevin Lin, Zi-ang Cao, Jingyun Yang, Rika Antonova, Marion Lepert, Sophie Lueth, Haoyu Xiong, Huy Ha, Cheng Chi, Philipp Wu, Fred Shentu, and Zhongke Yi for fruitful technical discussions. We thank Rajat Kumar Jenamani, Rishabh Madan, and the Cornell EmPRISE Lab for open sourcing their compliant controller for the Kinova arm. We would also like to thank Zi-ang Cao, Rika Antonova, Haoyu Xiong, Suvir Mirchandani, Matt Strong, Wesley Guo, and Claire Chen for extensive hardware assistance, as well as Stanford ILIAD, Stanford ARMLab, Rika Antonova, and Zipeng Fu for lending hardware. We are especially grateful towards the FIRST Robotics Competition (FRC) community for developing the rich ecosystem that made this project feasible. For extensive product and logistics support, we would like to thank Mandy Gove, Cory Ness, Omar Zrien, Jacob Caporuscio, and Dalton Smith from CTR Electronics; Ranjit Chahal and Harvey Rico from WestCoast Products; and John Rigsby from Swerve Drive Specialties. The work was supported in part by the Stanford Institute for Human-Centered Artificial Intelligence (HAI), Princeton School of Engineering, the Sloan Foundation, and the National Science Foundation under ECCS-2143601.

BibTeX

@inproceedings{wu2024tidybot,
  title = {TidyBot++: An Open-Source Holonomic Mobile Manipulator for Robot Learning},
  author = {Wu, Jimmy and Chong, William and Holmberg, Robert and Prasad, Aaditya and Gao, Yihuai and Khatib, Oussama and Song, Shuran and Rusinkiewicz, Szymon and Bohg, Jeannette},
  booktitle = {Conference on Robot Learning},
  year = {2024}
}