Trajectory planning by variable length chunk of sequence-to-sequence using hierarchical decoder

Abstract

Object manipulation is one of the most important issues for robots supporting humans. However, trajectory planning for dynamic manipulation is a difficult issue due to dynamic constraints. This paper deals with trajectory planning for dynamic manipulation using sequence-to-sequence (seq2seq) models. In a conventional method, multiple samples are chunked in order to reduce the length of the time series. However, as the chunk size is fixed, accuracy may be reduced and trajectories are truncated. Therefore, we verified the effects of these. We solved these problems by a seq2seq model using a hierarchical decoder. As a result, proposed model was able to generate a trajectory more accurate than the conventional method.