Event
MiTaS multi-resolution tactile framework achieves 80% success rate on five manipulation tasks
1 reports · 1 sources
Rickmer Krohn et al. propose MiTaS, a multi-resolution tactile perception framework that integrates information from an RGB camera, a GelSight Mini vision-based tactile sensor, and a high-frequency event-based Evetac sensor, conditioning a flow matching policy to perform contact-intensive manipulation tasks. The framework employs modality-specific convolutional stems and a Transformer-based fusion architecture. The authors report that in systematic evaluations on five contact-intensive manipulation tasks, MiTaS achieved an average success rate of 80%, whereas the vision-only baseline was 35.5%, and the visuotactile baselines were 55.5% and 63%, which were unable to reliably complete the tasks.
Report timeline
- MiTaS multi-resolution tactile imitation learning framework for contact-rich manipulation tasks
MiTaS is a multi-resolution tactile representation framework that uses modality-specific convolutional stems and a Transformer-based fusion structure to integrate information from an RGB camera, a GelSight Mini vision-based tactile sensor, and a high-frequency event-based Evetac sensor, then conditions a flow-matching policy to perform downstream tasks.
arXiv cs.RO ·