Robotic System From KIT Adapts to Disassemble Faulty Machines

Researchers at the Karlsruhe Institute of Technology (KIT) in Germany have developed a robotic system designed to disassemble machines even when parts are missing or components do not match the original design.

The system employs a probabilistic planning approach, specifically a Partially Observable Markov Decision Process (POMDP), to handle uncertainty and adapt its actions during the disassembly process. This allows the robot to update its plan in real time as it encounters unexpected situations, such as stuck screws or absent fasteners.

Experiments conducted by the research team included scenarios where the robot encountered a deliberately stuck screw in an electric motor. In response, the system switched from attempting to unscrew the fastener to using a milling tool to remove material and access the desired part. In another test involving an angle grinder, the robot detected that a screw was already missing and skipped the unnecessary unscrewing step.

The research was presented at the 2026 IEEE International Conference on Robotics and Automation (ICRA) in Vienna. Project designer Jan Baumgärtner of KIT introduced the system at the event, highlighting its ability to plan disassembly order, predict faults, and verify each action during operation.

What the numbers show

• In 30 manipulation trials, the robot removed the target part in every case.
• Screw removal succeeded 60% of the time without a screw-search step.
• With a screw-search step, screw removal success reached 100%.

According to reports, the system relies on computer-aided design (CAD) models and inspection data to predict defects and preserve valuable components during the disassembly process. This approach enables the robot to adapt its strategy if a machine behaves differently from its original design, such as when a fastener is stuck or missing.

Documentation from eWeek confirms that the robot can detect deviations from expected machine behavior and change tactics mid-task. For example, the robot can switch from unscrewing to milling when it encounters a stuck fastener, allowing it to continue the disassembly without manual intervention.

An arXiv preprint published in November 2025 describes the mathematical framework behind the system. The report states that the probabilistic planner outperformed deterministic approaches in both average disassembly time and consistency, and successfully adapted to missing or stuck parts across three products and two robotic systems.

IEEE Spectrum reports that the system is designed to preserve valuable parts while disassembling devices, using inspection data and CAD models to guide its actions. This method allows the robot to efficiently handle end-of-life products and recover components that might otherwise be lost due to damage or design changes.

* This article is based on publicly available information at the time of writing.