A team from Shanghai Jiao Tong University has launched an open-source project called U-Arm, which allows users to control 95% of robotic arms remotely for just 400 yuan. This innovative solution has already been validated on real robotic arms such as XArm6, Dobot CR5, and ARX R5.
Current mainstream data collection methods rely heavily on remote operation. However, many traditional remote operation systems, such as the ALOHA project, require identical robotic arms for operation and can cost over $20,000. On the other hand, lower-cost solutions like VR systems and game controllers face challenges such as singularities and compatibility issues.
The U-Arm project offers a cost-effective and adaptable solution called LeRobot-Anything-U-Arm, which can be assembled for about 400 yuan and operates with 95% of mainstream robotic arms. The system supports three configurations tailored to the most common robotic arm designs currently on the market.
As demonstrated in the images, remote operation verification has been conducted on the XArm6, Dobot CR5, and ARX R5 robotic arms. In terms of software compatibility, U-Arm supports a ROS-based control model that decouples command sending and receiving. Users simply need to subscribe to the angular joint topics published by U-Arm for different robotic arms, enabling seamless integration.
The hardware design of U-Arm is optimized for remote operation, with all elements being open-source and easily replicable. Unlike previous remote-controlled arms built on 3D printing with expensive servo mechanisms, U-Arm employs a new hardware solution that reduces costs while enhancing maintainability and durability. The servos used in the system are priced at only 45 yuan each, culminating in a total system cost (excluding 3D printing material) of under 400 yuan.
The design and assembly have also been optimized for user-friendliness during operation. For instance, the gearboxes of servos are removed in favor of encoders, ensuring that friction during joint movement is solely from adjustable screws. This smooth operation minimizes issues where the robotic arms may falter under gravity when nearing workspace limits.
Efficiency and data quality studies were conducted through five manipulation tasks using U-Arm, such as picking up soda cans from different shelf heights, stacking items, and sorting products. The U-Arm outperformed a standard game controller in average operation time by 39%, which is attributed to its optimized control architecture. However, in precision tasks, like stacking cans, U-Arm demonstrated lower success rates due to unanticipated movements by the operator, whereas the controller allowed them to act with greater precision.
Notably, U-Arm also achieved more natural movement trajectories compared to the game controller, benefiting from better distribution similarity during joint training, which aids model convergence.
The project has been fully open-sourced on GitHub, offering hardware STL and STEP files, software routines, and assembly guides, along with test routines for the SAPIEN simulation environment. The U-Arm collected data from the XArm6 is also available on Hugging Face. Project GitHub Link: github.com/MINT-SJTU/LeRobot-Anything-U-Arm Related Technical Report: arxiv.org/abs/2509.02437 Data Collected using U-Arm: https://huggingface.co/MINT-SJTU