Event-based 6-DoF Visual Servoing for Dynamic Objects

Visual servoing of dynamic objects remains challenging for conventional frame-based cameras due to motion blur and limited temporal resolution. Event cameras, with their microsecond-level latency and high dynamic range, are a natural fit for this regime, yet event-based 6-DoF visual servoing has not been demonstrated. We propose EDFT+, a tracker that evaluates both registration-based and prediction-based solvers on a single shared distance-field cost surface, interleaving frequent low-cost hypothesis-search steps with periodic Levenberg-Marquardt registration steps. Building on this, we present the first event-based 6-DoF visual servoing system for dynamic objects, coupling event-driven object tracking with a pose-based controller on a Franka Panda manipulator. We validate the system in a dual-robot setup, in which a second manipulator moves the target along diverse 6-DoF trajectories at speeds up to 0.4 m/s.

@article{kang2026event_servoing,
  title={Event-based 6-DoF Visual Servoing for Dynamic Objects},
  author={Kang, Yufan and Glover, Arren and Caron, Guillaume and Gava, Luna and Duvinage, Thomas and Ishikawa, Ryoichi and Oishi, Takeshi},
  journal={Under Review},
  year={2026}
}