TidyBot++: A Cost-Effective, Holonomic Mobile Manipulator for Robot Learning

Robotics research is currently experiencing rapid progress, especially in the field of imitation learning. To fully exploit the potential of these advancements, large amounts of demonstration data are necessary, collected through human interaction. TidyBot++ presents itself as a promising solution for this challenge. It is an open-source, holonomic mobile manipulator specifically designed for robot learning in the household context.

Cost-Effective and Flexible Design

TidyBot++ is characterized by its cost-effective and flexible design. The use of commercially available components and the open architecture allow researchers and developers to easily replicate the system and adapt it to their individual needs. A special feature is the compatibility with various robot arms, which enables a wide range of tasks.

Advantages of Holonomy

Unlike conventional mobile robots with differential drive, which are limited in their movement, TidyBot++ uses a holonomic base with driven caster wheels. This allows for independent and simultaneous control of all three degrees of freedom in the plane (X, Y translation and rotation). This significantly simplifies navigation in tight spaces and the execution of complex manipulation tasks. Time-consuming maneuvers required by non-holonomic systems are eliminated.

Intuitive Data Acquisition

Data acquisition for imitation learning is supported by an intuitive smartphone interface. Using the touchscreen and the IMU of the smartphone, users can remotely control the robot and record demonstrations for various tasks. This data then serves as the basis for training AI models.

Diverse Application Scenarios

Experiments have shown that TidyBot++ is capable of handling a variety of everyday tasks in the household. These include, for example, opening refrigerators, wiping surfaces, loading dishwashers, emptying trash cans, watering plants, and filling washing machines. The data collected within the project and the resulting AI models are available to the research community.

Further Developments and Outlook

The developers of TidyBot++ are continuously working on improving the system and expanding its capabilities. Future developments could include the integration of more advanced sensors, the improvement of autonomy, and the development of more robust learning algorithms. TidyBot++ has the potential to significantly advance research in the field of mobile manipulation learning and accelerate the development of intelligent robots for the household.

Additional Information

In addition to the hardware design and software for teleoperation, the developers of TidyBot++ also provide simulation environments and URDF files. This allows researchers to test and optimize algorithms in simulation before deploying them on the real robot. The documentation and source code are publicly available and promote collaboration and exchange within the research community.

Bibliographie: https://openreview.net/forum?id=L4p6zTlj6k https://tidybot2.github.io/ https://gfx.cs.princeton.edu/pubs/Wu_2024_TAO/index.php https://openreview.net/pdf?id=Kxy1VslITk https://tidybot2.github.io/poster.pdf https://www.corl.org/program/demos https://x.com/leto__jean https://www.youtube.com/watch?v=6MH7dhyIUdQ https://iprl.stanford.edu/publications https://github.com/jimmyyhwu/tidybot