Automating hand-eye calibration for robots

Camera-based robot systems are a central component of modern production and assembly processes today. In areas such as mechanical and plant engineering, automation, and medical technology, robots should be able to reliably assemble and disassemble components. To achieve this, robots and cameras have to be precisely calibrated to each other, similar to the interaction between the human eye and hand, otherwise the robot will miss its target.

State of the art

Until now, this so-called hand-eye calibration has often been done manually: operators move the robot to individual positions, capture image data, and use this to determine the exact position of the tool, such as a gripper or screwdriver. The achievable accuracy depends heavily on the experience of the operator and the selection of measurement points. External measuring systems such as laser trackers or coordinate measuring machines can provide very precise data, but they are associated with high costs.

Technology

Researchers from the Digital Process Engineering (DPE) group at the Institute for Mechanical Process Engineering and Mechanics (MVM) at KIT have successfully developed a method that automates and mathematically optimizes hand-eye calibration. The system is built on a digital model of the robot and camera that precisely describes their movement and positional relationships to each other. Based on this, an algorithm calculates a continuous motion path (trajectory) optimized for calibration, which the robot then follows autonomously. In the real-world demonstrator, a camera continuously records the position of the tool via an optical marker on the tool, while the robot simultaneously logs the tool position. The system uses this synchronous data to calculate the exact spatial relationship between the camera, robot, and tool. Additionally, the system statistically evaluates the accuracy achieved.

Advantages

The process is fully automatic and delivers precise and reproducible results – regardless of the operator's experience. It is suitable for both stationary and robot-mounted camera systems. It can also be integrated into existing systems and enables the retrofitting of existing robots for new visual applications.

Options for companies

A demonstrator is currently under construction. KIT is looking for partners for application-specific development projects and licensing of the technology.