Handling Ball Spin and Complex Serves with Egocentric Onboard Perception

Develop reactive perception–control strategies, potentially conditioned on opponent behavior, that enable a humanoid table-tennis robot relying solely on egocentric onboard cameras to robustly handle ball spin and complex serves, despite the current lack of explicit spin modeling and the difficulty of observing spin from visual input alone.

Background

The paper introduces SMASH, a humanoid table-tennis system that achieves agile whole-body striking using only egocentric onboard perception. While the system demonstrates robust performance across diverse strikes, it deliberately omits explicit modeling of ball spin.

The authors note that spin is challenging to infer from visual observations, making complex serves and spin handling difficult under the current design. They explicitly state that addressing spin—likely via opponent-conditioned reactive strategies—remains an open challenge, highlighting a key limitation for advancing autonomous, onboard-perception-based humanoid table tennis.