【24h】

Learning Bicycle Stunts

机译:学习自行车特技

获取原文
获取原文并翻译 | 示例
           

摘要

We present a general approach for simulating and controlling a humanrncharacter that is riding a bicycle. The two main componentsrnof our system are offline learning and online simulation. We simulaternthe bicycle and the rider as an articulated rigid body system.rnThe rider is controlled by a policy that is optimized through offlinernlearning. We apply policy search to learn the optimal policies,rnwhich are parameterized with splines or neural networks for differentrnbicycle maneuvers. We use Neuroevolution of AugmentingrnTopology (NEAT) to optimize both the parametrization and the parametersrnof our policies. The learned controllers are robust enoughrnto withstand large perturbations and allow interactive user control.rnThe rider not only learns to steer and to balance in normal riding situations,rnbut also learns to perform a wide variety of stunts, includingrnwheelie, endo, bunny hop, front wheel pivot and back hop.
机译:我们提出了一种模拟和控制骑自行车的人类角色的通用方法。我们系统的两个主要组件是离线学习和在线仿真。我们将自行车和骑手模拟为铰接式刚体系统。骑手受通过离线学习优化的策略控制。我们应用策略搜索来学习最优策略,该策略通过样条或神经网络进行参数化,以实现不同的自行车机动。我们使用AugmentingrnTopology(NEAT)的神经进化来优化我们的策略的参数化和参数。博学的控制器足够坚固耐用,可以承受较大的扰动并允许交互式用户控制。骑手不仅学会在正常骑行情况下进行转向和平衡,而且还学会执行各种特技,包括轮式,内胎式,兔子跳,前轮式枢轴和后跳。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号