German researchers unveil robot that adapts to dismantle faulty machines

Researchers at Germany's Karlsruhe Institute of Technology have developed a robotic disassembly platform that can cope with the uncertainties of worn or altered products. Beginning with a design model, the robot inspects each part, revises its internal map when movements differ from expectations, and assigns probabilities to possible faults. This Partially Observable Markov Decision Process approach lets the machine choose alternative actions, like using a milling tool when a screw is seized or skipping a missing fastener.

Experiments on an electric motor and an angle grinder demonstrated that the probabilistic planner outperformed rigid plans as the likelihood of stuck components increased. The findings were shared at the 2026 IEEE International Conference on Robotics and Automation in Vienna. While the current demos focus on small devices, the team envisions scaling the technology to larger factories to enable more circular manufacturing and cheaper automated repairs.