SGD for robot motion? the effectiveness of stochastic optimization on a new benchmark for biped locomotion tasks

Martim Brandao, Kenji Hashimoto, Atsuo Takanishi

    研究成果: Conference contribution

    1 引用 (Scopus)

    抄録

    Trajectory optimization and posture generation are hard problems in robot locomotion, which can be nonconvex and have multiple local optima. Progress on these problems is further hindered by a lack of open benchmarks, since comparisons of different solutions are difficult to make. In this paper we introduce a new benchmark for trajectory optimization and posture generation of legged robots, using a pre-defined scenario, robot and constraints, as well as evaluation criteria. We evaluate state-of-The-Art trajectory optimization algorithms based on sequential quadratic programming (SQP) on the benchmark, as well as new stochastic and incremental optimization methods borrowed from the large-scale machine learning literature. Interestingly we show that some of these stochastic and incremental methods, which are based on stochastic gradient descent (SGD), achieve higher success rates than SQP on tough initializations. Inspired by this observation we also propose a new incremental variant of SQP which updates only a random subset of the costs and constraints at each iteration. The algorithm is the best performing in both success rate and convergence speed, improving over SQP by up to 30% in both criteria. The benchmark's resources and a solution evaluation script are made openly available.

    元の言語English
    ホスト出版物のタイトル2017 IEEE-RAS 17th International Conference on Humanoid Robotics, Humanoids 2017
    出版者IEEE Computer Society
    ページ39-46
    ページ数8
    Part F134101
    ISBN(電子版)9781538646786
    DOI
    出版物ステータスPublished - 2017 12 22
    イベント17th IEEE-RAS International Conference on Humanoid Robotics, Humanoids 2017 - Birmingham, United Kingdom
    継続期間: 2017 11 152017 11 17

    Other

    Other17th IEEE-RAS International Conference on Humanoid Robotics, Humanoids 2017
    United Kingdom
    Birmingham
    期間17/11/1517/11/17

    Fingerprint

    Biped locomotion
    Quadratic programming
    Robots
    Trajectories
    Learning systems
    Costs

    ASJC Scopus subject areas

    • Artificial Intelligence
    • Computer Vision and Pattern Recognition
    • Hardware and Architecture
    • Human-Computer Interaction
    • Electrical and Electronic Engineering

    これを引用

    Brandao, M., Hashimoto, K., & Takanishi, A. (2017). SGD for robot motion? the effectiveness of stochastic optimization on a new benchmark for biped locomotion tasks. : 2017 IEEE-RAS 17th International Conference on Humanoid Robotics, Humanoids 2017 (巻 Part F134101, pp. 39-46). IEEE Computer Society. https://doi.org/10.1109/HUMANOIDS.2017.8239535

    SGD for robot motion? the effectiveness of stochastic optimization on a new benchmark for biped locomotion tasks. / Brandao, Martim; Hashimoto, Kenji; Takanishi, Atsuo.

    2017 IEEE-RAS 17th International Conference on Humanoid Robotics, Humanoids 2017. 巻 Part F134101 IEEE Computer Society, 2017. p. 39-46.

    研究成果: Conference contribution

    Brandao, M, Hashimoto, K & Takanishi, A 2017, SGD for robot motion? the effectiveness of stochastic optimization on a new benchmark for biped locomotion tasks. : 2017 IEEE-RAS 17th International Conference on Humanoid Robotics, Humanoids 2017. 巻. Part F134101, IEEE Computer Society, pp. 39-46, 17th IEEE-RAS International Conference on Humanoid Robotics, Humanoids 2017, Birmingham, United Kingdom, 17/11/15. https://doi.org/10.1109/HUMANOIDS.2017.8239535
    Brandao M, Hashimoto K, Takanishi A. SGD for robot motion? the effectiveness of stochastic optimization on a new benchmark for biped locomotion tasks. : 2017 IEEE-RAS 17th International Conference on Humanoid Robotics, Humanoids 2017. 巻 Part F134101. IEEE Computer Society. 2017. p. 39-46 https://doi.org/10.1109/HUMANOIDS.2017.8239535
    Brandao, Martim ; Hashimoto, Kenji ; Takanishi, Atsuo. / SGD for robot motion? the effectiveness of stochastic optimization on a new benchmark for biped locomotion tasks. 2017 IEEE-RAS 17th International Conference on Humanoid Robotics, Humanoids 2017. 巻 Part F134101 IEEE Computer Society, 2017. pp. 39-46
    @inproceedings{857a78517dd441a08ba0a8d2c7539d71,
    title = "SGD for robot motion? the effectiveness of stochastic optimization on a new benchmark for biped locomotion tasks",
    abstract = "Trajectory optimization and posture generation are hard problems in robot locomotion, which can be nonconvex and have multiple local optima. Progress on these problems is further hindered by a lack of open benchmarks, since comparisons of different solutions are difficult to make. In this paper we introduce a new benchmark for trajectory optimization and posture generation of legged robots, using a pre-defined scenario, robot and constraints, as well as evaluation criteria. We evaluate state-of-The-Art trajectory optimization algorithms based on sequential quadratic programming (SQP) on the benchmark, as well as new stochastic and incremental optimization methods borrowed from the large-scale machine learning literature. Interestingly we show that some of these stochastic and incremental methods, which are based on stochastic gradient descent (SGD), achieve higher success rates than SQP on tough initializations. Inspired by this observation we also propose a new incremental variant of SQP which updates only a random subset of the costs and constraints at each iteration. The algorithm is the best performing in both success rate and convergence speed, improving over SQP by up to 30{\%} in both criteria. The benchmark's resources and a solution evaluation script are made openly available.",
    author = "Martim Brandao and Kenji Hashimoto and Atsuo Takanishi",
    year = "2017",
    month = "12",
    day = "22",
    doi = "10.1109/HUMANOIDS.2017.8239535",
    language = "English",
    volume = "Part F134101",
    pages = "39--46",
    booktitle = "2017 IEEE-RAS 17th International Conference on Humanoid Robotics, Humanoids 2017",
    publisher = "IEEE Computer Society",

    }

    TY - GEN

    T1 - SGD for robot motion? the effectiveness of stochastic optimization on a new benchmark for biped locomotion tasks

    AU - Brandao, Martim

    AU - Hashimoto, Kenji

    AU - Takanishi, Atsuo

    PY - 2017/12/22

    Y1 - 2017/12/22

    N2 - Trajectory optimization and posture generation are hard problems in robot locomotion, which can be nonconvex and have multiple local optima. Progress on these problems is further hindered by a lack of open benchmarks, since comparisons of different solutions are difficult to make. In this paper we introduce a new benchmark for trajectory optimization and posture generation of legged robots, using a pre-defined scenario, robot and constraints, as well as evaluation criteria. We evaluate state-of-The-Art trajectory optimization algorithms based on sequential quadratic programming (SQP) on the benchmark, as well as new stochastic and incremental optimization methods borrowed from the large-scale machine learning literature. Interestingly we show that some of these stochastic and incremental methods, which are based on stochastic gradient descent (SGD), achieve higher success rates than SQP on tough initializations. Inspired by this observation we also propose a new incremental variant of SQP which updates only a random subset of the costs and constraints at each iteration. The algorithm is the best performing in both success rate and convergence speed, improving over SQP by up to 30% in both criteria. The benchmark's resources and a solution evaluation script are made openly available.

    AB - Trajectory optimization and posture generation are hard problems in robot locomotion, which can be nonconvex and have multiple local optima. Progress on these problems is further hindered by a lack of open benchmarks, since comparisons of different solutions are difficult to make. In this paper we introduce a new benchmark for trajectory optimization and posture generation of legged robots, using a pre-defined scenario, robot and constraints, as well as evaluation criteria. We evaluate state-of-The-Art trajectory optimization algorithms based on sequential quadratic programming (SQP) on the benchmark, as well as new stochastic and incremental optimization methods borrowed from the large-scale machine learning literature. Interestingly we show that some of these stochastic and incremental methods, which are based on stochastic gradient descent (SGD), achieve higher success rates than SQP on tough initializations. Inspired by this observation we also propose a new incremental variant of SQP which updates only a random subset of the costs and constraints at each iteration. The algorithm is the best performing in both success rate and convergence speed, improving over SQP by up to 30% in both criteria. The benchmark's resources and a solution evaluation script are made openly available.

    UR - http://www.scopus.com/inward/record.url?scp=85044436842&partnerID=8YFLogxK

    UR - http://www.scopus.com/inward/citedby.url?scp=85044436842&partnerID=8YFLogxK

    U2 - 10.1109/HUMANOIDS.2017.8239535

    DO - 10.1109/HUMANOIDS.2017.8239535

    M3 - Conference contribution

    AN - SCOPUS:85044436842

    VL - Part F134101

    SP - 39

    EP - 46

    BT - 2017 IEEE-RAS 17th International Conference on Humanoid Robotics, Humanoids 2017

    PB - IEEE Computer Society

    ER -