Scribd
Upload
108 views1 page

E08 Planning PDF

The document provides instructions for exercises involving path planning using rapidly-exploring random trees (RRTs) and defining a distance metric in phase space for a 1D point mass system. Specifically, it asks the student to: 1) Grow an RRT backward from a target configuration and forward from the origin to find a collision-free path, then smooth the path. 2) Define a distance metric in phase space for a point mass system that respects the dynamics, quantify the distance between two states, and use it to grow an RRT toward a target state.

Uploaded by

Praveen D Jadhav
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
108 views1 page

E08 Planning PDF

The document provides instructions for exercises involving path planning using rapidly-exploring random trees (RRTs) and defining a distance metric in phase space for a 1D point mass system. Specifically, it asks the student to: 1) Grow an RRT backward from a target configuration and forward from the origin to find a collision-free path, then smooth the path. 2) Define a distance metric in phase space for a point mass system that respects the dynamics, quantify the distance between two states, and use it to grow an RRT toward a target state.

Uploaded by

Praveen D Jadhav
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

Robotics

Exercise 8
Marc Toussaint
Lecturer: Duy Nguyen-Tuong
TAs: Philipp Kratzer, Janik Hager, Yoojin Oh
Machine Learning & Robotics lab, U Stuttgart
Universitätsstraße 38, 70569 Stuttgart, Germany

December 17, 2018

1 RRTs for path finding (8 points)


1. Please update the version of the repository by running the following.

git pull
git submodule update
make -C rai dependAll
make -j4

2. For python you can run: ’jupyter-notebook py/06-rrt/06-rrt.ipynb’

3. For C++ run: ’cd cpp/06-rrt’, ’make’, ’./x.exe’

The code demonstrates an RRT exploration that randomly samples from Q and displays the explored endeffector
positions.
a) First grow an RRT backward from the target configuration q ∗ = (0.945499, 0.431195, −1.97155, 0.623969, 2.22355,
−0.665206, −1.48356). Grow the RRT directly towards q = 0 with a probability of β = 0.5 . Stop when there exists
a node close (<stepSize) to the q = 0 configuration. Read out the collision free path from the tree and display it.
Why would it be more difficult to grow the tree forward from q = 0 to q ∗ ? (3 points)
b) Find a collision free path using bi-directional RRTs (that is, 2 RRTs growing together). Use q ∗ to root the backward
tree and q = 0 to root the forward tree. Stop when a newly added node is close to the other tree. Read out the
collision free path from the tree and display it. (3 points)
c) Propose a simple method to make the found path smoother (while keeping it collision free). Implement the method
and display the smooth trajectory. (2 points)

2 A distance measure in phase space (4 points)


Consider the 1D point mass with mass m = 1 state x = (q, q̇). The 2D space of (q, q̇) combining position and velocity
is also called phase space. In RRT’s (in line 5 on slide 05:39) we need to find the nearest configuration qnear to a
qtarget . But what does “near” mean in phase space? This exercise explores this question.
Consider a current state x0 = (0, 1) (at position 0 with velocity 1). Pick any random phase state xtarget ∈ R2 .
How would you connect x0 with xtarget in a way that fulfils the differential constraints of the point mass dynamics?
Given this trajectory connecting x0 with xtarget , how would you quantify/meassure the distance? (If you defined the
connecting trajectory appropriately, you should be able to give an analytic expression for this distance.) Given a set
(tree) of states x1:n and you pick the closest to xtarget , how would you “grow” the tree from this closest point towards
xtarget ?

You might also like