This action might not be possible to undo. Are you sure you want to continue?

A fascinating example of using an I-map to dramatically reduce the I-space was given a long time ago by Blum and Kozen [121]. Map building requires space that is linear in the number of tiles; however, it is possible to ensure that the environment has been systematically searched using much less space. For 2D grid environments, the searching problem can be solved without maintaining a complete map. It must systematically visit every tile; however, this does not imply that it must remember all of the places that it has visited. It is important only to ensure that the robot does not become trapped in an infinite loop before covering all tiles. It was shown in [121] that any maze can be searched using space that is only logarithmic in the number of tiles. This implies that many different environments have the same representation in the machine. Essentially, an I-map was developed that severely collapses space. down to a smaller derived I-

Assume that the robot motion model is the same as given so far in this section; however, no map of the environment is initially given. Whatever direction the robot is facing initially can be declared to be north without any harm. It is assumed that any planar 2D grid is possible; therefore, there are identical maps for each of the four orientations. The north direction of one of these maps might be mislabeled by arbitrarily declaring the initial direction to be north, but this is not critical for the coming approach. It is assumed that the robot is a finite automaton that carries a binary counter. The counter will be needed because it can store values that are arbitrarily large, which is not possible for the automaton alone. To keep the robot from wandering around in circles forever, two important pieces of information need to be maintained:

1. The latitude, which is the number of tiles in the north direction from the robot's initial position. 2. When a loop path is executed, it needs to know its orientation, i.e., whether the loop travels clockwise or counter clockwise.

Both of these can be computed from the history I-state, which takes the same form as in (13.12), except in the current setting, is given by (13.23), and is the set of all bounded environments (bounded means that the white tiles can be contained in a large rectangle). From the history I-state, let denote the

subsequence of the action history that corresponds to actions that produce motions. The latitude, , can be computed by counting the number of actions that produce motions in the north direction and subtracting those that produce motions in the south direction. The loop orientation can be determined by angular odometry (which is equivalent to having a compass in this problem [287]). Let the value give the number of right turns in yields a clockwise loop, and minus the number of left turns in . Note that making four rights

. Making four lefts yields a counterclockwise loop, and

. In general, it can be shown that for any loop path that does not intersect itself, either , which means that it travels clockwise, or counterclockwise. , which means that it travels

Assuming that the robot can always detect a unique point along the boundary. then the robot traveled counterclockwise. and there is an outer boundary. imagine that the obstacles are protruding from the walls. the robot moves around the boundary of all obstacles in a single closed loop path.It was stated that a finite automaton and a binary counter are needed. If there is some special object in the environment. There is a fictitious thin obstacle that extends southward from each unique point. The unique point is defined as the westernmost vertex among the southernmost vertices of the obstacle. This is the power of working directly with an I-space over the set of environments. then the robot traveled clockwise. By refusing to cross these fictitious obstacles. If the result is . In this sense. contrary to the case of maintaining latitude. To solve the problem. even though at the end. The strategy so far does not ensure that every cell will be visited.14. Alternative plans do not need to be computed from the problem data because the plan can handle all possible environments without modification. Therefore. as opposed to ``floating'' in the interior. then the given strategy is always guaranteed to find it. it can imagine that the obstacles are connected as shown in Figure 13. The plan will be described in terms of the example shown in Figure 13.15. which can be detected when reached by the robot. as opposed to requiring state estimation. a unique point can be determined on the boundary of each obstacle.15.6. This observation avoids using an unlimited number of bits. By using the latitude and orientation information. the modification shown in Figure 13. The robot can use its sensing information to move in a way that keeps the wall to its left. Using the latitude and orientation information. Blum and Kozen were the ``planner'' that found a plan that solves any problem. For any environment. the robot moves to a boundary and traverses it by performing wall following. the unique point can always be found (see Exercise 4). which assumed a complete map was given. The counter is used to keep track of as the robot moves. it does not even have a map! The resulting approach can be considered as an information-feedback plan on the I-space. In other words. It is interesting to compare the solution to the spanning-tree coverage planning approach in Section 8. . If the result is after forming a loop. and on the outer boundary.16 is needed to ensure that every tile is visited by zig-zag motions. This connects the obstacles together in a way that appears to be an extension of the outer boundary. These are shown by small discs in Figure 13. It turns out that an additional counter is not needed to measure the angular odometry because the robot can instead perform mod-3 arithmetic when counting right and left turns. The construction so far can be viewed as part of an I-map that maps the history I-states into a much smaller derived I-space. there are obstacles in the interior (this example has six). and the goal was to optimize the distance traveled.

Figure 13.14: An example that has six obstacles. b) An alternative loop that visits all of the tiles in the interior. LaValle 2005-12-23 . Figure 13.15: The obstacles are connected together by extending a thin obstacle downward from their unique points. Next: Stentz's Algorithm (D) Up: Grid Problems Previous: Algorithms for navigation Contents Index Steven M. Figure 13.16: a) A clockwise loop produced by wall following.

Sign up to vote on this title

UsefulNot useful- Lv Gave Too Different for Comfort
- Flow_medium Pressure Function
- robots vs humanos
- tro_2013
- Accurate Indoor Position Estimation Aim
- Product Specification 4400 M98A BWOS3.2
- S4C-Product Manual IRB 4400 3HAC 2932-1 Rev.2 M98
- Cloud in Robo
- Robotic
- MobilePhone Controlled Robot Complete Document
- Robot
- Error_code_Fanuc
- Robot Smile
- Sensing Bot
- Carrier Prog 300708
- Plejadian Selauset
- robowarz regulation
- Unit 4 - Bhs Inggris
- rOBOTICS NOTES
- School Robotics
- Autonomous workshop
- Pedoman Penulisan Proposal KRI2008 (Wahidin W.)
- ECE.pdf
- An Untethered Miniature Origami Robot
- TOWARD ROBOTS AS SOCIALLY AWARE PARTNERS IN HUMAN-ROBOT INTERACTIVE TASKS
- Proposal for Li2-Robo-III Workshop
- Individual Assignment
- 2002 Solarbotics solarspeeder
- 0417_y16_sm_1
- Kinect Based Autonomous and Gesture Controlled Robo Soccer
- Algorithms for Maze Searching

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue reading from where you left off, or restart the preview.