Embedded Motion Control 2014 Group 4

Members of group 4

Khy, R.N.	Norent
Kiriouchine, V.I.	Seva
Kuijk, F.J.M. van	Fransis
Marx, S.	Sanne
Smit, J.	Jorrit

Media

Find videos of the Pico emc04 tests on our Youtube channel.
See here an animated GIF of simulated Pico, where he takes a corner and stops: Animation

Planning

Week 1:
-Introduction Lecture
-Determine course goals

Week 2:
-Install Ubuntu, ROS, fix SVN etc.: get everything up and running
-Finish all tutorials: C++, ROS
-Think of first concepts to tackle the corridor problem
-Implement first concepts: wallfinder, angle/distance controller

Week 3:
-Improve angle/distance controller
-Try to implement holonomic concept
-Meet tutor to discuss problems and project progress
-Perform Pico Test

Week 4:
-[Sanne] Place current functions in seperate .cpp files to allow for parallel updating
-[Sanne]Change the sendVelocity function into a twoPointNavigator function which aims to place pico between two lazer scan ranges+angles given as function input.
-[Fransis]Create a corridorFinder function which identifies the two corners of all corridors large enough for pico to pass through. Discard distant corridors and output two lazer scan ranges+angles of the closest found corridor.
-[Seva]Create a function which decides what points are sent to the twoPointNavigator function based on how close pico is to the side corridor found by corridorFinder , allowing pico to seamlessly switch between navigating the straight corridor and turning into the side corridor.
-[Norent]Change the safety radius into a safety box around pico using the dimensions of pico, create an unStuck function which allows pico to safely move away from any obstacle breaching the safety box.
-Implement a finishedCorridor condition which can detect that pico has left the corridor exit and stops all movement.
-[Pico EMC04]Win the corridor competition.

Concepts

wallFinder
Pico should be able to recognize walls. This can be done by using the Laser Range Finder. The data received from the Laser Range Finder must be filtered to get useful data that can be used by our controller.
From this sensor we receive an array with distances, 270 degrees around Pico. This array is split in half to represent vision on the left and right side. In these two arrays we search for the smallest distance that is greater than 0.15 m (a part of the vision is blocked by Pico him/her/it-self). Finally the angle at wich these distances are with repsect to Pico are calculated. This function thus finds two distances and two corresponding angles.

Angle/Distance controller
When Pico navigates through the corridor the distances found by the wallFinder are controlled to be equal. Besides the distances, the angles at which these distances are detected should be controlled to be zero. When the closest distances left and right of pico are equal and the angles zero, Pico is oriented in the middle of the corridor oriented in the longitudinal direction, which is exactly what we want.

Holonomic movement
At the tutor meeting it became clear that Pico is equipped with omni-wheels which means holonomic movement is possible. Currently the Pico in the simulator has differential wheels. With omni-wheels the distance can be controlled independent of the angle.

Cornering
We should think of how Pico can recognize and take corners.

findExit
We should think of how Pico can recognize the exit of the corridor so that it stops moving.

safetyHitbox
TODO hitbox dimenions (front x side): 2.0e-1 x 1.5e-1.

unstuckSafety
TODO

Implementation

Wallfinder
The wallfinder is called with every lazerscan callback and will return the nearest scanpoint on the left and the nearest scanpoint on the right of pico.

Corridorfinder
The corridorfinder is called with every lazerscan callback and will return the closest corner point of the nearest corridor and the furthest corner point of the closest corridor to pico.
The closest corner is found by looking at the scan data forward of the found wallpoints and determining whether the difference between two subsequent points is large enough. The furtherst corner of the corridor is found by looking forward again past the closest point. When a local minimum is found in the scan distance this is called the furthest corner of the corridor. If the difference between the closest and furthest corner is not large enough for pico to pass through, the corridor is considered a gap.

Masterplanner
The masterplanner determines where to navigate. If the closest corridor corner is very close to one of the points found by the wallfinder, the pico will take the corner using the turnAroundPoint function. If no side corridor is found or the corridor is not close enough, pico will drive forward.
If while taking the corner the closest point on the other side of pico is found at an angle much different from +-90 degrees, the masterplanner knows the pico is mid corner. As soon as the wallfinder again finds the points left and right of pico at the same angle the masterplanner will switch to driving forward.

twoPointNavigator
The twoPointNavigator function will send the velocitys to pico in order to keep pico moving forward between parallel walls. The distance of closest points left and right of pico is controlled to be equal by controlling the linear.y speed. The angle of pico in relation to the walls is controlled by sending an angular.z speed such that the closest point left and right are at and equal angle from the centre of pico.

turnAroundPoint
The turnAroundPoint function will send the velocitys to pico in order to keep pico moving forward while keeping a constant distance to one of the wallfinder points, maintaining this point at a +/- 90 degree angle w.r.t. pico depending on what side the side corridor is.

stopFinal
The stopFinal function is called at each lazercallback to see if the pico has exited the maze using two checks. The first check checks whether all lazer range points are larger than a certain distance. This is the minimum distance pico needs to travel in order for it to consider itself outside of the maze. The second check splits up the lazer scan into 10 sectors, calculating the average distance for each sector. The average distance of each sector has to be greater than a certain (larger than the for the first used check) value. This ensures that the pico will stop close to the maze exit when the majority of the surroundings are very far away.

geometryIdentifier
The geometryIdentifier identifies whether pico is close to any corridors, intersections or junctions, it replaces the corridorFinder for the final test.
The steps taken by the geometryIdentifier are given in pseudo code as:

if (found corridor on the left)

then

if (found corridor on the right)

then

if (left and right corridor are at equal distance to pico

then

if (furthest points of both corridors are wide enough to be a corridor)

then -> Pico has found an intersection

else -> Pico has found a three way left right junction

else -> Pico has found two seperate corridors

else

if (furthest corridor point is much closer than point in front of pico)

then -> Pico has has found a three way left forward junction

else -> Pico has found a corridor to the left

else

if (found corridor on the right)

then

if (furthest corridor point is much closer than point in front of pico)

then -> Pico has found a three way right forward junction

else -> Pico has found a corridor to the right

else -> Pico is in a corridor with no side corridors

Arrow recognition

The output of the arrow recognition is as given in table below.

message	output
nothing	0
left	1
right	2

First the RGB image is transformed to HSV and afterwards red-filtered to a binary image. Note the following:

imshow of RGB image looks the way it should.

imshow of HSV image does not, this is because imshow is not meant for HSV.

red-filtering from hue 170 to 10 still needs to be done right. currently 170 to 180 is used.

Three tools available for image recognition:

Floodfill

Line detection

Corner detection

Corner detection does not seem to be useful in its current form (as given in the slides), an improved version of corner detection might solve this problem. Also the parameters for corner detection might not be tuned well enough.

Log

Week 2:
Monday 28-04-2014: Group meeting SEL/SOL
Present: Sanne, Seva, Norent, Fransis

• Installed all software on laptops
  
• Got the svn up and running
  
• Started with tutorials and discussed problems
  

Wednesday 30-04-2014: Meeting SEL/SOL
Present: Sanne, Fransis

• Finished tutorials
  

Friday 02-05-2014: Meeting OGO 2
Present: Sanne, Seva, Norent, Fransis

• Created concepts: wallFinder, angle/distance controller, exitFinder, Cornering
  
• Implemented the following concepts: wallFinder, angle/distance controller:
  
• Pico now can detect walls and navigate through the corridor, not controlled optimal yet

Week 3:
Wednesday 07-05-2014: Group meeting OGO 18
Present: Sanne, Seva, Norent, Fransis

• Held first tutor meeting with Sjoerd
  
• Created holonomic concept
  
• Prepared for real world Pico test
  

Thursday 08-05-2014: First experiment with Pico
Present: Sanne, Seva, Norent, Fransis

• Preparation for the experiment by reviewing the tutorials and our work
  
• Tested code on Pico
  
• Saved logfiles
  
• Observed Pico's behaviour as expected:
  
  • Safety precautions work
    
  • Speed limits work
    
  • Robust for gaps in wall
    
  • Robust for small corridors (width: 8e-1 [m])
    
  • Robust for non-straight corridors
    
  • Unable to make the turn for the corridor contest
    
• Measured Pico rectangular hitbox:
  
  • front: 1.5e-1 [m]
    
  • side: 2.0e-1 [m]
    
  • Made videos of experiments (uploaded soon™)

missing a part

Thursday 28-05-2014: Work session
Present: Sanne, Seva, Norent, Fransis

• Worked in seperate groups on decision making and arrow recognition
  
• Meeting with Sjoerd
  

  Time Table

  Overview of time spent per group member

  	Lectures	Group meetings	Mastering ROS and C++	Preparing midterm assignment	Preparing final assignment	Wiki progress report	Other activities
  Week_1_Norent	2
  Week_1_Seva	2		6
  Week_1_Fransis
  Week_1_Sanne	2		2
  Week_1_Jorrit
  Week_2_Norent	2		6
  Week_2_Seva	2	6	2
  Week_2_Fransis	2	4	6
  Week_2_Sanne	2	6	4			0.5
  Week_2_Jorrit
  Week_3_Norent		1	3	3			1
  Week_3_Seva	2	12
  Week_3_Fransis	2	8
  Week_3_Sanne	2	12
  Week_3_Jorrit
  Week_4_Norent		13					2
  Week_4_Seva		15					2
  Week_4_Fransis		15
  Week_4_Sanne		15					2
  Week_4_Jorrit
  Week_5_Norent
  Week_5_Seva
  Week_5_Fransis
  Week_5_Sanne
  Week_5_Jorrit
  Week_6_Norent
  Week_6_Seva
  Week_6_Fransis
  Week_6_Sanne
  Week_6_Jorrit
  Week_7_Norent
  Week_7_Seva
  Week_7_Fransis
  Week_7_Sanne
  Week_7_Jorrit
  Week_8_Norent
  Week_8_Seva
  Week_8_Fransis
  Week_8_Sanne
  Week_8_Jorrit
  Week_9_Norent
  Week_9_Seva
  Week_9_Fransis
  Week_9_Sanne
  Week_9_Jorrit
  Week_10_Norent
  Week_10_Seva
  Week_10_Fransis
  Week_10_Sanne
  Week_10_Jorrit