2019-08-07

ICRA2019_papers_reading

Num title author time
1 Dynamic Channel: A Planning Framework for Crowd Navigation Chao Cao1, Peter Trautman2 and Soshi Iba2 2019_CMU
2 Deep Reinforcement Learning of Navigation in a Complex and Crowded Environment with a Limited Field of View Guillaume Sartoretti1, Justin Kerr1, Yunfei Shi1, Glenn Wagner2, 2019
3 Crowd-Robot Interaction:Crowd-aware Robot Navigation with Attention-based Deep Reinforcement Learning Changan Chen, Yuejiang Liu, Sven Kreiss and Alexandre Alahi VITA, Ecole Polytechnique Federal de Lausanne, EPFL, Switzerland 2019
4 PRIMAL: Path-finding via Reinforcement and Imitation Multi-Agent Learning T. K. Satish Kumar3, Sven Koenig3, and Howie Choset1 2019
5 Deep Reinforcement Learning Robot for Search and Rescue Applications: Exploration in Unknown Cluttered Environments Farzad Niroui, Student Member, IEEE, Kaicheng Zhang, Student Member, IEEE, Zendai Kashino, Student Member, IEEE, and Goldie Nejat, Member, IEEE 2019

①- Dynamic Channel: A Planning Framework for Crowd Navigation

摘要

  1. 动态环境导航中考虑时间维度会严重消耗算力
  2. 局部规划器只处理即刻的碰撞,不能长线优化
  3. 联合高维拓扑路径规划和低维运动规划的新方法,三角空间内的图搜索问题

介绍

  • 现有工作关于局部或者全部最优,局部规划器只寻找短时间内的免碰撞路径而忽略了全局最优性。
  • 引入动态通道,一种全新的将全局最优和启发式相结合的拥挤导航框架,高效处理智能体动态。
  • 使用三角分割抽象环境,再用改进的A*算法计算最优路径。

局部和全局规划

  传统局部路径规划方法有DWA(Dynamic Window Approach), IVS(Inevitable Collision States), and VO(Velocity Obstacles), VOVO(Variants of Velocity Obstacles)。都基于多智能体有相同的导航策略,但不能保证生成最短移动路径。
  全局规划,基于采样算法有RRT(Rapidly-exploring Random Tree)和PRM(Probabilistic roadmap methods),规划是认为环境是静态,再规划是再考虑动态。
通过层次规划和潜在领域的结合,可以实现静态和动态障碍的分离。

  1. 常用Voronoi图(维诺图)和Delaunay triangulation(三角分割),相比栅格地图搜索更小的图,且考虑了拓扑属性,如连接性。TA* \ TRA*

你可能感兴趣的:(2019-08-07)