Robotik 1 - Kapitel 7 - Bewegungsplanung (Teil 3)
A HumanoidRobots lecturer explains sampling based robot motion planning, narrow passage strategies, available software and how planned trajectories feed robot control.
In this talk
- 0:02
From graph search to sampling based planners
The lecture reviews planning from a start configuration to a goal, beginning with graph construction and A*. It covers Probabilistic Roadmaps for repeated queries in static environments, Dynamic Roadmaps, RRTs, bidirectional RRTs, constraint aware RRTs and RRT*, which can improve paths by reconnecting nearby nodes at lower cost.
- 4:22
Why narrow passages challenge RRTs
Uniform sampling can spend many samples in open areas while missing tight openings, where collision free routes may be difficult to find. Dynamic Domain RRT narrows sampling around boundary nodes, approximating visible Voronoi regions with n dimensional balls of radius R. In the examples, bidirectional Dynamic Domain RRT solved a difficult case in 1.6 seconds, while standard RRT found no solution after 80,000 seconds.
- 18:29
Bridge sampling targets tight openings
Bridge sampling first seeks a colliding configuration, then samples a second nearby point in the collision region. If the midpoint between them is collision free, the planner uses it to extend the tree. The approach can also guide local connections in a Probabilistic Roadmap, concentrating samples near narrow passages.
- 21:58
Planning tools and next steps
The lecturer recommends checking the Open Motion Planning Library, or OMPL, for existing sampling based planners, noting its C++ and Python implementations. CMOX is presented as a lighter C++ platform for robot simulation, motion planning and grasp planning. The chapter closes by linking collision free paths from planning to desired control values, and introduces grasp planning as the next topic.
Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.