Beruflich Dokumente
Kultur Dokumente
Part XI
Motion Planning
Department of Informatics
Intelligent Autonomous Systems
Outline
Motion Planning
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Outline
Motion Planning
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Motion Planning
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Motion Planning
http://planning.cs.uiuc.edu/
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Given obstacles, a robot, and its motion capabilities compute collision-free robot
motions from the start to goal.
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Geometric Models
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
O = H1 H2 H3 H4 .
and from finite unions of those:
O = H1 H2 H3 H4 .
O could then become any semi-algebraic set.
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Whats inside?
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Department of Informatics
Intelligent Autonomous Systems
Department of Informatics
Intelligent Autonomous Systems
Department of Informatics
Intelligent Autonomous Systems
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
For the case of two links, C = S 1 S 1 , the obstacle region already becomes
strange and complicated!
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Department of Informatics
Intelligent Autonomous Systems
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Department of Informatics
Intelligent Autonomous Systems
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Completeness Notions
forever.
Resolution complete: If a solution exists at a given resolution, it finds
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Combinatorial Planning:
geometry
All algorithms are complete
Usuall produce a roadmap in Cfree
Extremely efficient for low-dimensional problems but dont scale well for
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Department of Informatics
Intelligent Autonomous Systems
Department of Informatics
Intelligent Autonomous Systems
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Three cases:
Department of Informatics
Intelligent Autonomous Systems
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Sampling-Based Planning
Explicit construction of the obstacle space is often intractable!
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Sampling: Denseness
In topology, a set U is called dense in V if cl(U) = V .
Implication: Every open subset of V contains at least one point in U.
If U is dense and countable, a dense sequence can be formed:
:NU
Example: The rational numbers Q are dense in R.
Department of Informatics
Intelligent Autonomous Systems
The set of all limit points of U is a closed set called the closure of U, and is
denoted by cl(U). Note: cl(U) = int(u) U.
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Halton sequence: For each coordinate, use relatively prime bases. More uniform
than random (which needs O((lg n)1/d ) as many samples to produce the same
expected dispersion.
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Sampling: Dispersion
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
1
1
2bk d c
Department of Informatics
Intelligent Autonomous Systems
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Rapidly Exploring Random Trees (RRT) are one variant of Rapidly Exploring
Dense Trees (RDT), others may use a non-random dense sequence.
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
within radius .
Among all paths to the root from xnew , add a new RRT edge only for the
shortest one.
If possible to reduce cost for other vertices within radius by connecting to
xnew , then disconnect them from their parents and connect them through
xnew .
The radius is prescribed through careful percolation theory analysis
(related to dispersion)
RRT yield asymptotically optimal paths through Cfree .
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Connection Rules:
Nearest K: The K closest points to (i) are considered.
Component K: Try to obtain up to K closest points of each connected
component of G.
Radius: Take all points within a ball of radius r centered at (i).
Visibility: Try connecting (i) to all vertices in G.
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Due to robot kinematics and dynamics, most systems are locally constrained in
addition to global obstacles.
Let q represent the C-space velocity.
In ordinary planning, any direction is allowed and the magnitude does not
matter.
Thus we could say
q = u
and u Rn may be any velocity vector.
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
More generally, a control system (or state transition equation) constrains the
velocity:
q = f (q, u)
and u belongs to some set U (usually bounded).
A function u : T U is applied over a time interval T = [0, tf ] and the
configuration q(t) at time t is given by
Z
q(t) = q(0) +
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
C = R2 S 1 .
Let u = (us , u ) and U = [0, 1] [max , max ].
Control system of the form q = f (q, u) :
x = cos
y = sin
us
=
tan u
L
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Department of Informatics
Intelligent Autonomous Systems
Manipulation Planning
In most forms of motion planning, the robot is not allowed to touch objects!
In manipulation planning, two phases are distinguished:
Transit mode: The robot moves towards a part.
Transfer mode: The robot carries a part.
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
We distinguish further:
Base Transit: The base moves to bring the object into reachability of the
arms.
Pregrasp: The gripper is brought into a pose from where it can go straight.
Grasping: The gripper goes straight to the grapsing pose and closes.
Carry Pose: The object is moved into a specific place close to the robot to
facilitate navigation.
Base Transit: The base moves to bring the goal into reachability.
Placing: The object is placed at the goal pose.
Arm Retract: Similar to the Pregrasp, we move the gripper out in a
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Department of Informatics
Intelligent Autonomous Systems
Octomaps are used to represent the environment, as they are they have a small
memory footprint and allow for hierarchical queries.
Often, we want to encode free space, obstace space and unknown space.
Therefore, 2 bits per node are needed.
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Whenever an object is carried, its model gets attached to the robot model.
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
the environment?
Arm planning: Can we find a trajectory via planning or can we execute the
Department of Informatics
Intelligent Autonomous Systems
The robot is sliced vertically and convex hulls for each slice are used to speed up
collision checks.
We can then use RRT or any other planner as usual.
Motion Planning
Michael Beetz
Summer Term 2012
Department of Informatics
Intelligent Autonomous Systems
Exam Question
Motion Planning
Michael Beetz
Summer Term 2012