کنفرانس مهندسی برق ایران

صفحه اصلی / سی و چهارمین کنفرانس بین المللی مهندسی برق

An Autoencoder-Based Approach to Generalizable Slip Detection for Robotic Grasping

نویسندگان :

Aysan Alizadeh Arasi¹ Ali Sadighi²

1- دانشگاه تهران 2- دانشگاه تهران

کلمات کلیدی :

vision-based tactile sensor،slip detection،force estimation،autoencoder،grip force control

چکیده :

Developing generalizable slip detectors for adaptive robotic grasping is challenging, as deep learning models often overfit object-specific features and require diverse labeled data. To this end, we adopt a two-stage framework in which an autoencoder first learns features from unlabeled data, and a recurrent network is then trained on top of the frozen encoder using limited labeled data to learn temporal slip patterns. The resulting Encoder-LSTM model achieves competitive predictive performance relative to established architectures such as ConvLSTM, while offering significantly lower computational cost. The shared encoder also supports force estimation, allowing both tasks to use a single feature representation suitable for embedded deployment. The system's practical viability was validated in a closed-loop experiment, demonstrating that the robot can maintain a stable grasp using tactile feedback alone.