Comp150-07: Intelligent Robotics
Lab 6: Localization
Due Tuesday, March 3, in class

We are building up to Project 1 where we build a robot capable of navigating an unknown maze and locating a goal landmark. To achieve that capability, the robot must be able to map its environment, localize itself in it, and plan a path from its location to the goal or a waypoint. In previous labs, we assumed a correct map and initial localization, and we worked on path planning. Now we relax our assumptions to allow realistic, imperfect sensors, and we learn to localize using that noisy sensory information.

The following two lab parts are not equal in scope and difficulty. Moreover, results from part 1 (Sensor characteristics) will be directly used in part 2 (Localization). So you should plan on work on both of them as a team. This lab is time-consuming. Start early!

Part 1: Sensor characteristics

Test #1: Distance

1. What is the variance of your sensor?
2. Do you have any problems with outliers or noisy readings at any of the points? Document them.
3. How will you deal with outlier and noisy readings issues when using this sensor for localization?
4. Extra credit: What happens if you face a wide obstacle at a non-90-degree angle to the sonar? Give a suggestion for interpreting sonar readings when you don't know which way an obstacle may be facing.

Test #2: Angle of perception

5. What is the error and variance of your sensor at different angles? What does the sensor's perception cone look like?
6. Do you have any outliers or particularly noisy readings? Document them.
7. Extra credit: What happens if you face a wide obstacle at a non-90-degree angle to the sonar? Is there a qualitative difference between behavior you've observed in question 4? Why?

Part 2: Localization

During demo on Tuesday, we will place your sonar-equipped robot at an undisclosed position (within sonar range) and orientation somewhere in a 250cm x 125cm workspace, in which you know there are 3 "point-like" obstacles (they will be small-footprint but tall enough to be sensed) at positions (50,50), (100,50), and (50,100). The robot's objective is to determine its initial pose in the workspace.

Rules:

• The robot may drive and/or turn as you like, but must return to its original pose if it does.
• It must then display its initial pose in global world coordinates on the screen.

Hand in: Your report should contain

• Design description related to sensing, including a labelled picture. Why did you choose to place the sensor(s) as you did?
• Discussion of early failure modes and obviation.
• High-level pseudo-code or flowchart description of your localization algorithm.
• Heavily commented code (post online and send me a link, send a file by email or use the provide system - don't print this part out for the hardcopy).

Tips and things to think about

2. In the first part of the lab, you discover the properties of your sonar sensors' response over a range of distances and angles. This response should be fairly consistent most of the time, but it's likely to contain regions of weirdness and/or occasional outliers.

Depending on the level of noise you measure, you may not need to perform any filtering on your sonar signal (a low-pass filter is already implemented inside the LEGO sensor). If you do encounter noise, you can implement a simple linear filter such as a moving window average: you keep a small buffer of recent values and treat their average as your sensor value instead of the raw value.

3. Your localization algorithm will need to disregard the outliers (if any) while making use of good information. You can tell outliers apart by noticing that they are significantly different from the mean of nearby/recent values. Since you know the variance of your sensor from part 1 of this lab, you can choose to discard any value that is different from the mean by more than 2 &sigma or 3 &sigma (&sigma² = variance). However, be sure to make a distinction between outliers and a legitimate change in sensor readings (e.g., because the obstcale face has ended and there is nothing in front of the sensor anymore).

4. You will most probably need to use active sensing to localize, since you only have 2 sonars. Active sensing means: use robot motion and odometry information to get distance estimates from 2 or more robot positions and orientations, then compute original pose from this information.