HIGHLEVELMANIPULATION PRIMITIVESFORAROBOTARM Supported by National - - PowerPoint PPT Presentation
ThesisProposalfortheMasterofScienceDegreeinComputerScience PresentedbyGlennV.Nickens AdvisedbyDr.DavidTouretzky Thursday,January29,2009 HIGHLEVELMANIPULATION
Thesis Proposal for the Master of Science Degree in Computer Science Presented by Glenn V. Nickens Advised by Dr. David Touretzky Thursday, January 29, 2009
1
Supported by National Science Foundation awards 0717705 and 0742106 to Carnegie Mellon University, and 0742198 to Norfolk State University.
2
computer science undergraduate programs due to:
– The complexity of the underlying mathemaKcs and theory that is needed to implement kinemaKcs, path planning and manipulaKon – The size, cost and/or limited capabiliKes of available robots
Mellon University have developed:
– The Tekkotsu soSware framework – An inexpensive hand‐eye robot ($995 from RoPro Design)
3
– is an open source robot programming framework based on C++. – is an object‐oriented and event passing architecture that makes full use of the template and inheritance features of C++. – programmers use high‐level primiKves, such as “nod your head” and “walk to that locaKon”, to control robots.
4
– has a three‐link planar arm, with a webcam a`ached to a pan‐Klt above it for vision. – is a small, affordable robot that fits on a classroom table. – can be programmed to manipulate objects idenKfied by the camera.
5
6
– design the high‐level manipulaKon primiKves required to program the hand‐eye robot. – formalize and implement the algorithms for the primiKves. – demonstrate that the primiKves are suitable for performing complex tasks.
7
– A robot arm – KinemaKcs – Collision detecKon – Path planning – Path smoothing
8
servos
are constructed of aluminum tubes and elongated c‐brackets
c‐bracket
9
pushing them
surfaces:
– Interior of the c‐bracket – Wrist – Forearm & Upper arm
10
plane requires at least 3DOF (x, y, and θ).
2DOF.
reach a target from many angles; a two‐link arm can reach it from
11
link arm will have at most two soluKons, a four‐link arm may have infinitely many soluKons.
four‐link arm is more complex than for a three‐ link arm.
(power, weight, wiring).
12
the relaKonship between the posiKon and orientaKon of the end‐effector and the joint angles of the arm.
13
14
calculaKons search for a set of joint angles that produce a specified end effector configuraKon.
calculaKons are harder; there may be mulKple soluKons, or there may be none.
15
16
Determine the: 1) wrist posiKon [xw, yw] from the target point [xt, yt] and desired orientaKon, ϕt. 2) elbow angle, θ2, from the wrist posiKon [xw, yw] and arm dimensions, L1 and L2. 3) shoulder angle, θ1. 4) wrist angle, θ3.
17
posiKon [xw, yw] from the target point [xt, yt] and desired orientaKon, ϕt.
18
angle, θ2, from the wrist posiKon [xw, yw] and arm dimensions, L1 and L2.
19
determining the elbow angle, θ2, may produce two soluKons.
elbow up and elbow down configuraKons.
20
angle, θ1.
21
angle, θ3
22
23
full 360° because adjacent arm segments will collide.
configuraKons that are within the turning limits
24
but may be reached from another orientaKon.
intervals for 180° in both direcKons unKl a valid soluKon is found.
25
Desired
Valid
are valid, the IK solver must choose one.
as the arm’s current configuraKon.
26
detecKon algorithm to check for collisions of the arm with itself or with
are represented as rectangles.
the SeparaKng Axis Theorem.
27
Once the kinemaKcs solver has determined the start and end configuraKons for the arm, a collision free path must be computed to move the arm.
28
29
Random Trees (RRTs) are trees whose verKces encode configuraKon states of the arm (LaValle 1998).
alternately extending the tree in random direcKons and moving toward the goal configuraKon.
30
RRTs, one from the start and one from the goal configuraKon, and biases the trees to grow toward each other.
the path is extracted using backtracking.
31
backtracking from the two trees.
32
33
the arm’s trajectory along a path from the start (green) to the end (red) configuraKon.
constructed by the path planner.
same path, but aSer the smoothing algorithm has been applied to it.
34
– The high‐level primiKves I have chosen so far – How I plan on implemenKng them and – How I will demonstrate the success of my work
35
36
37
38
39
40
– Which manipulaKon surface to use – How to acquire contact with an object – What trajectory the arm should follow
41
arm’s freedom of moKon.
with the upper arm or the red obstacle if it tries to move inward or
42
43
Planned trajectories must keep the manipulation surface in contact with the object
Moving an object with the upper arm or the forearm is imprecise because it may slide.
44
Some tasks may require mulKple manipulaKons
45
In this sequence of images, the arm first moves the blue object with its forearm, then moves it with the c-bracket to get it to the target position.
– Monitor the execuKon of each step – Take correcKve acKon should something go wrong
46
– verifying arm posiKons by checking the joint angles. – verifying object posiKons using vision.
– The arm may not have acquired the object. – The object may not be where it should be. – The arm may have accidentally moved the object.
47
– Moving a game piece – Sweeping game pieces off the board – Picking a game piece from a group
48
49
manipulate objects by specifying desired effects and levng the planner figure out how to achieve them.
able to experiment with manipulaKon without having to write their own kinemaKcs solvers, path planners, or collision detecKon algorithms.
50
planner and engine will be integrated into the Tekkotsu framework and will be used on other robots, including the Chiara.
51
52
53
Tekkotsu web site: http://tekkotsu.no-ip.org/gettingstarted.html.
and Control. John Wiley & Sons, 2006.
http://www.kuffner.org/james/plan/. January 13, 2009.
to single-query path planning. In Proceedings IEEE International Conference on Robotics and Automation, pages 995--1001, 2000. http://msl.cs.uiuc.edu/~lavalle/rrtpubs.html.
http://msl.cs.uiuc.edu/~lavalle/rrtpubs.html, Oct. 1998.
54