A Comparison of Imitation Learning Algorithms for Bimanual Manipulation

Michael Drolet, Simon Stepputtis, Siva Kailas, Ajinkya Jain, Jan Peters, Stefan Schaal, Heni Ben Amor

Keywords: Bimanual Manipulation, Imitation Learning, Robotics

TL;DR: We extensively compare prominent imitation learning algorithms on a bimanual manipulation platform.

Abstract: Amidst the wide popularity of imitation learning algorithms in robotics, their properties regarding hyperparameter sensitivity, ease of training, data efficiency, and performance have not been well-studied in high-precision industry-inspired environments. In this work, we explore the limitations and advantages of prominent imitation learning algorithms and evaluate them on a complex bimanual manipulation task involving multiple contacts. We show that while imitation learning is effective for such tasks, not all algorithms are equal in handling environmental and hyperparameter perturbations, training demands, and usability. Our study uses a carefully designed experimental procedure to assess these key characteristics.

Submission Number: 18