Google's DeepMind and Stanford have launched an enhanced version of ALOHA - ALOHA 2. Compared to the first generation, ALOHA 2 has enhanced performance, ergonomic design and robustness, and costs less than 200,000 RMB. And, in order to accelerate the research of large-scale two-handed operation, all hardware designs related to ALOHA 2 are open sourced, and detailed tutorials are provided, as well as the ALOHA 2 MuJoCo model with system recognition function. Google's DeepMind released the paper "ALOHA 2: An Enhanced Low-Cost Hardware for Bimanual Teleoperation".

• Address: https://aloha-2.github.io/assets/aloha2.pdf

• Project Homepage: https://aloha-2.github.io/

Let's take a look at what the upgraded ALOHA 2 can do, such as putting different toys into three different bowls.

Juggling, you throw me to pick up.

Open the Coke bottle and pour the Coke into another cup, open the yogurt box.

Put contact lenses on your panda doll.

What's even more unimaginable is that it can turn into a pickpocket, silently take your wallet, and put it back for you.

It's absolutely stunning! ALOHA 2 dramatically increases the durability of the first generation of ALOHA, enabling large-scale data collection on more complex tasks.

Compared to the first generation, ALOHA 2 has been upgraded

One of the goals is to scale up data collection on the ALOHA platform to support the study of complex operational tasks, including the number of robots used, the number of data collection hours per robot, and the diversity of data collection. This scaling process changes the requirements and scope relative to the first-generation ALOHA platform.

For ALOHA 2, in addition to building on the ALOHA platform, the investigators are also looking for further improvements in the following areas:

• Performance & Mission Scope: Key components that enhance ALOHA's performance, including grippers and controllers, enable a wider range of manipulative tasks.

• User-friendliness and ergonomics: To optimize large-scale data collection, prioritize user experience and comfort, including improving the responsiveness and ergonomics of user interface systems.

• Robustness: Increase the robustness of the system and minimize downtime due to diagnostics and repairs. This requires simplifying the mechanical design and ensuring that the larger fleet of robots is generally easy to maintain.

Based on the above objectives, the specific improvements of ALOHA 2 are as follows:

• Grippers: The researchers designed a new low-friction track for the grippers of the master/slave robots. For the main robot, this improves the ergonomics and responsiveness of the teleoperation. For follower robots, this improves the delay and the force output of the gripper. In addition, they have upgraded the gripping tape material on their fingers to improve durability and the ability to grip small objects.

• Gravity compensation: The researchers created a passive gravity compensation mechanism using off-the-shelf components, which improved durability compared to ALOHA's original gripping material system.

• Frame: The researchers simplified the frame around the work cell while maintaining the rigidity of the camera mounting points. These changes provide space for props for human-robot collaborators and robots to interact.

• Camera: ALOHA 2 uses a smaller Intel RealSense D405 camera and a custom 3D printed camera mount to reduce the footprint of the follower arm, which reduces the obstruction of operational tasks. These cameras also have a larger field of view, depth, global shutter, and more customization features.

• Simulation: The researchers simulated the precise specifications of the ALOHA 2 robot in the MuJoCo model in MuJoCo Menagerie, resulting in improved data collection, strategy learning, and simulation evaluation for challenging manipulation tasks.

Grippers

In order to make remote control operation smoother and improve ergonomics, a low-friction track design has been adopted to reduce mechanical complexity, thus replacing ALOHA's original scissor guide manipulator design.

The researchers designed and built a low-friction follow-up manipulator to replace ALOHA's original design. The lower friction reduces the perceived delay between the lead robot and the follower robot gripper, significantly improving the user experience during remote operation.

frame

The researchers redesigned the support frame and made it out of 20x20mm aluminum profiles. The frame provides support for the lead robot and gravity compensation system, as well as mounting points for top-down and bug-eye cameras.

Compared to ALOHA, the design has been simplified by removing the vertical frame on the opposite side of the table and the remote operator. The increased space allows for more diverse ways of data collection. For example, human collaborators can more easily stand across from the workspace and interact with bots, collecting data on human-machine interactions. In addition, larger props can be placed in front of the workbench for the robot to interact with.

mock

The researchers published the MuJoCo Menagerie model for the ALOHA 2 work cell, which is very useful for remote operation and simulation learning.

MuJoCo's improved physical accuracy and visual fidelity compared to previously published ALOHA models allows for fast, intuitive, and scalable simulation data collection.

MuJoCo model rendering.

Simulate remote operation tasks.

Here's an example of using the Google Scanned Objects Dataset to perform remote operations with a MuJoCo model (1x faster):

Source: Heart of the Machine

It is only used for academic sharing, if there is any infringement, please leave a message and delete the infringement immediately!

This article is from Xinzhi self-media and does not represent the views and positions of Business Xinzhi.If there is any suspicion of infringement, please contact the administrator of the Business News Platform.Contact: system@shangyexinzhi.com