In Lab 10, we wrote the localization code by ourself to learn the processes, filters, and models used to localize a mobile robot. In the current lab, we are provided with an optimized implementation which we will modify to drive our physical robot. The optimized code utilizes precaching, vector processing, and other programming practices which significantly speed up the calculation time. To test this code out, I ran the simulation once again and made sure it yielded the expected results. Below is the output of the simulation, where The green line shows the robot's actual ground truth position, the red line is estimation based on odometry/sensor values, and the blue line is the Bayes filter estimation of the position.

Before we can run the simulation code on the real robot, we have to implement the perform_observation_loop() function of class RealRobot in lab11_real.ipynb. This code allows the robot to rotate at a uniform angular speed of 15 degrees per second and collect 18 evenly spaced time of flight sensor readings each 20 degrees apart. This array of 18 data points is then passed into the optimized localization program to provide the final estimation of the robot's location. Here is the code I used to implement this functionality:

Using the new localization code, I placed my robot at the various marked poses in the lab and ran the routine to take 18 sensor measurements and predict the robot's location. The results for each location are as follows:

(0,0)

Although this pose is not one of the 4 included in the lab, my localization result matched perfectly with the ground truth, so I had to include it :)

(-3,-2) and (0,3)

I was also able to get pretty good results for the above two points after many tries. The green dot represents the ground truth position, and the blue dot is the localization result.

(5,3)

This point was harder to localize, and I was able to get mixed results after many tries. However, there were many runs where I successfully collected and constructed a polar plot of the 18 sensor data points just to have the localization process return a completely wrong prediction with high confidence. One change that helped the localization process to return the correct prediction was to manually remove some data points which should be farther away but somehow recorded to be much closer to the robot due to faulty ToF sensor readings. After making these points appear farther away as expectd, the localization performed better.

(5,-3)

Lastly, it was very difficult for me to localize this point. After many attempts, the localization process kept mistaking (5,-3) for a similar point farther to the left. Since both points have a wall to the right and bottom sides and are more open on the other sides, the localization process may have mistaken these two wall features. In addition, my ToF sensors are not extremely reliable and sometimes returned faulty values or sensed objects that weren't there which made it hard to accurately localize.

In summary, my robot's ability to localize is inconsistent and unreliable. Even for the poses that yielded more or less successful localization results, I had to run many trials in order to achieve sufficiently valid data. I discussed this problem with the TAs in the lab and learned that it is not uncommon for ToF sensors to make unreliable measurements, especially at longer distances. I had the same problem in the stunt lab where my robot would often detect a wall without there being one in front of the robot, and it would prematurely stop and attempt the stunt when it was actually very far from the wall. This issue seemed to happen again in this lab, and since there are only 18 data points used in the localization process, even one faulty data point can cause the prediction to be erraneous. This is in contrast to the mapping lab, where I was able to achieve very accurate results with my ToF sensor despite its occasional flawed data readings due to the sheer volume of data I was collecting. In conclusion, due to difficulty with my ToF sensors and getting accurate sensor readings, my robot was unable to consistently localize correctly. Therefore, I do not plan to use localization in Lab 12, as I am afraid it will not yield consistent and accurate results.