Deep Robot Path Planning from Demonstrations for Breast Cancer Examination

In 2020, breast cancer affected around two million people worldwide. Early cancer detection is, therefore, needed to save many lives and reduce treatment costs. Nowadays, mammography and self- palpation are the most popular monitoring methods. The high number of cases and the difficulty of correct self-diagnosis has prompted this research work to design a fully autonomous robot to perform breast palpation. Specifically, this study focuses on learning the path for a successful breast examination of a silicone model. Learning from demonstrations proved to be the most suitable approach to reproduce the desired path. We implemented a teleoperation control between two Franka Emika Panda robots with tactile and force feedback to perform palpation on both simple and complex shapes. Moreover, we created a dataset of simple palpation strategy. Finally, we developed and tested different sequential neural networks such as Recurrent Neural Network (RNN), Long short-term memory (LSTM), Gated recurrent unit (GRU) and Temporal Convolutional Network (TCN) to learn the stochastic behaviour of the acquired palpation trajectories. The results showed that TCN is capable of reproducing the desired behaviour with more accuracy and stability than the other models.