TY - GEN
T1 - A Real-Robot Dataset for Assessing Transferability of Learned Dynamics Models
AU - Agudelo-Espana, Diego
AU - Zadaianchuk, Andrii
AU - Wenk, Philippe
AU - Garg, Aditya
AU - Akpo, Joel
AU - Grimminger, Felix
AU - Viereck, Julian
AU - Naveau, Maximilien
AU - Righetti, Ludovic
AU - Martius, Georg
AU - Krause, Andreas
AU - Scholkopf, Bernhard
AU - Bauer, Stefan
AU - Wuthrich, Manuel
N1 - Publisher Copyright:
© 2020 IEEE.
PY - 2020/5
Y1 - 2020/5
N2 - In the context of model-based reinforcement learning and control, a large number of methods for learning system dynamics have been proposed in recent years. The purpose of these learned models is to synthesize new control policies. An important open question is how robust current dynamics-learning methods are to shifts in the data distribution due to changes in the control policy. We present a real-robot dataset which allows to systematically investigate this question. This dataset contains trajectories of a 3 degrees-of-freedom (DOF) robot being controlled by a diverse set of policies. For comparison, we also provide a simulated version of the dataset. Finally, we benchmark a few widely-used dynamics-learning methods using the proposed dataset. Our results show that the iid test error of a learned model is not necessarily a good indicator of its accuracy under control policies different from the one which generated the training data. This suggests that it may be important to evaluate dynamics-learning methods in terms of their transfer performance, rather than only their iid error.
AB - In the context of model-based reinforcement learning and control, a large number of methods for learning system dynamics have been proposed in recent years. The purpose of these learned models is to synthesize new control policies. An important open question is how robust current dynamics-learning methods are to shifts in the data distribution due to changes in the control policy. We present a real-robot dataset which allows to systematically investigate this question. This dataset contains trajectories of a 3 degrees-of-freedom (DOF) robot being controlled by a diverse set of policies. For comparison, we also provide a simulated version of the dataset. Finally, we benchmark a few widely-used dynamics-learning methods using the proposed dataset. Our results show that the iid test error of a learned model is not necessarily a good indicator of its accuracy under control policies different from the one which generated the training data. This suggests that it may be important to evaluate dynamics-learning methods in terms of their transfer performance, rather than only their iid error.
UR - http://www.scopus.com/inward/record.url?scp=85092744986&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85092744986&partnerID=8YFLogxK
U2 - 10.1109/ICRA40945.2020.9197392
DO - 10.1109/ICRA40945.2020.9197392
M3 - Conference contribution
AN - SCOPUS:85092744986
T3 - Proceedings - IEEE International Conference on Robotics and Automation
SP - 8151
EP - 8157
BT - 2020 IEEE International Conference on Robotics and Automation, ICRA 2020
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2020 IEEE International Conference on Robotics and Automation, ICRA 2020
Y2 - 31 May 2020 through 31 August 2020
ER -