Timed-ESTD 导航 | 20231005个人简介

Project Description

这个项目旨在解决高动态环境下的导航问题，特别是在密集的人群场景下；
该项目提出了一种基于梯度的规划期，通过state-time空间在线生成轨迹；
该项目提出了一种基于时间的ESDT参数，支持距离和梯度查询，用来优化轨迹；定义了状态-时间空间兼容的smooth prior和obstacle likelihood的似然函数并优化；
该项目使用模拟和基准数据集对该方法进行了验证；
更多的信息可以参考 this paper.

State-Time Graph Building:
- Graph Definition : Instead of searching in pixel level, we define the timed triangle graph, a topological representation of moving obstacles. It is composed of edge and face, which depend on pedestrians' positions and velocities.
- Dual Graph: We discrete the time dimension into T slices with a resolution dt. We then simplify eachi triangle into a node and build feasible connections for adjacent nodes.
- State-time Graph Building: We use the optimization method to execute the node-placement procedure when doing the searching.
State-Time A*:
- Once the time budget T is used up or the searching process fials, the current optimal path is returned.
- The neghtor retrival, i.e. SUCC is based on the graph, so the connections across diffrent time slices are built up.
- The node placement is performed along the progress of the path searching.

Prior Dsitribution of Smooth Trajectories:
- To parameterize smooth trajectories, a GP prior is defined over the trajectory space, and each trajectory is sampled from this prior distribution.
- Suppose a near-optimal time assignment t0< t1< t2 < ... < tf, is found by the front end, then the trajectory is discretized into F+1 waypoints accordingly.
Timed ESDF:
- ESDF is frequently used cost map, shown in Fig2, which contains distance and gradient information against obstacles.
- We defined the likelihood function to evaluate the clearance of the trajectory
MAP Trajectory:
- Based on GP prior and likelihood function, the back-end trajectory optimization os completed by solving a MAP problem, in equation (17)