Artificial Intelligence for Prosthetics - challenge solutions
Abstract
In the NeurIPS 2018 Artificial Intelligence for Prosthetics challenge, participants were tasked with building a controller for a musculoskeletal model with a goal of matching a given time-varying velocity vector. Top participants were invited to describe their algorithms. In this work, we describe the challenge and present thirteen solutions that used deep reinforcement learning approaches. Many solutions use similar relaxations and heuristics, such as reward shaping, frame skipping, discretization of the action space, symmetry, and policy blending. However, each team implemented different modifications of the known algorithms by, for example, dividing the task into subtasks, learning low-level control, or by incorporating expert knowledge and using imitation learning.
Cite
@article{arxiv.1902.02441,
title = {Artificial Intelligence for Prosthetics - challenge solutions},
author = {Łukasz Kidziński and Carmichael Ong and Sharada Prasanna Mohanty and Jennifer Hicks and Sean F. Carroll and Bo Zhou and Hongsheng Zeng and Fan Wang and Rongzhong Lian and Hao Tian and Wojciech Jaśkowski and Garrett Andersen and Odd Rune Lykkebø and Nihat Engin Toklu and Pranav Shyam and Rupesh Kumar Srivastava and Sergey Kolesnikov and Oleksii Hrinchuk and Anton Pechenko and Mattias Ljungström and Zhen Wang and Xu Hu and Zehong Hu and Minghui Qiu and Jun Huang and Aleksei Shpilman and Ivan Sosin and Oleg Svidchenko and Aleksandra Malysheva and Daniel Kudenko and Lance Rane and Aditya Bhatt and Zhengfei Wang and Penghui Qi and Zeyang Yu and Peng Peng and Quan Yuan and Wenxin Li and Yunsheng Tian and Ruihan Yang and Pingchuan Ma and Shauharda Khadka and Somdeb Majumdar and Zach Dwiel and Yinyin Liu and Evren Tumer and Jeremy Watson and Marcel Salathé and Sergey Levine and Scott Delp},
journal= {arXiv preprint arXiv:1902.02441},
year = {2019}
}