Gokhan Solak tames robot vibrations
- Gokhan Solak and Arash Ajoudani developed a control method that lets collaborative robots learn and cancel vibrations from handheld power tools while they work, targeting contact-heavy jobs such as polishing. - The method extends the band-limited multiple Fourier linear combiner with adaptive damping, and the authors report better suppression than recursive least squares and Kalman-based variants in simulations and polishing tests. - The work builds on a 2023 International Conference on Robotics and Automation paper and a 2025 arXiv preprint from the Italian Institute of Technology. (arxiv.org)
Robots can hold a vibrating power tool, but the shaking can throw off the job. Gokhan Solak and Arash Ajoudani built a controller that learns that shaking online and pushes back against it while the robot is working. (arxiv.org) The problem shows up in contact tasks, where the robot is pressing on a surface instead of moving through empty air. A drill, grinder, or similar tool adds fast oscillations that can reduce accuracy and increase wear. (zenodo.org) (arxiv.org) Solak and Ajoudani’s approach uses a band-limited multiple Fourier linear combiner, a learning algorithm that tracks repeating vibrations by breaking them into frequencies. The controller then applies a feedforward force, meaning it injects a counter-force before the vibration fully throws the robot off course. (arxiv.org) The newer paper adds what the authors call a damped BMFLC method. It changes the learning step size with a logistic damping rule so the system can react quickly without amplifying sensor noise. (arxiv.org) That matters for collaborative robots, which are designed to stay compliant around people instead of simply becoming rigid. The 2023 conference paper said many earlier fixes suppressed vibration by stiffening the robot, which can hurt performance when safe, flexible interaction is required. (zenodo.org) The experiments were done on a 7-degree-of-freedom Franka Panda arm carrying a modified handheld drill. In the 2025 preprint, the tool was fitted with an abrasive head to polish a wood block, and the drill speed was controlled through an Arduino-USB interface. (zenodo.org) (arxiv.org) The authors report that the damped method improved suppression rates over the original BMFLC and over recursive least squares and Kalman filter extensions in simulation. They also write that it was more computationally efficient than those two alternatives. (arxiv.org) This line of work has been moving from conference results toward a fuller journal-style preprint. Google Scholar lists “Online Learning and Suppression of Vibration in Collaborative Robots with Power Tools” at the 2023 IEEE International Conference on Robotics and Automation, and the expanded arXiv version was submitted on August 5, 2025. (scholar.google.com) (arxiv.org) Solak’s affiliation in both versions is the Italian Institute of Technology in Genoa, where Ajoudani is a co-author on several contact-rich robotics projects. The pitch is straightforward: let robots keep a light touch while handling tools that naturally want to shake them off target. (arxiv.org) (scholar.google.com)
