ENGN8536 Statistical Inference in Mechatronics - 2022
Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit
Problem Set 1
ENGN8536 Statistical Inference in Mechatronics - 2022
Objective and Submission Instructions
● The objective of this problem set is to practice applying Bayes’ theorem and hidden Markov model filtering.
● Students may collaborate to solve the problems but must write (handwrite or type) and submit their own individual solutions.
● Solutions are to be submitted as a single pdf file through the Turnitin link on Wattle.
● Late submissions are not permitted. Submissions after the due date without an extension will be awarded a mark of 0.
1 Bayes’Theorem (5 marks)
An autonomous car has detected an object. The car now needs to classify the object as one of three possible classes: a cyclist, a pedestrian, or a truck. Because the car is driving on a freeway, the car’s initial belief is that the probability of the object being a cyclist is 0.1, the probability of the object being a pedestrian is 0.1, and the probability of the object being a truck is 0.8. The car has a camera system running an object classifier that is known to have the following performance characteristics:
● If the object is actually a cyclist, the camera classifies it as: a cyclist with probability 0.6; a pedestrian with probability 0.3; and a truck with probability 0.1.
● If the object is actually a pedestrian, the camera classifies it as: a cyclist with probability 0.2; a pedestrian with probability 0.7; and a truck with probability 0.1.
● If the object is actually a truck, the camera classifies it as: a cyclist with probability 0.3; a pedestrian with probability 0.1; and a truck with probability 0.6.
If the camera system classifies the object as a cyclist, what are the probabilities (according to Bayes’ theorem) that the car should assign to the object being a cyclist, a pedestrian, and a truck?
2 Bayes’Theorem for Sensor Fusion (5 marks)
Suppose that the car in Problem 1 also has a lidar system that it can also use to classify the object. The lidar system is running a different object classifier that is known to have the following performance characteristics:
● If the object is actually a cyclist, the lidar detects it as: a cyclist with probability 0.5; a pedestrian with probability 0.3; and a truck with probability 0.2.
● If the object is actually a pedestrian, the lidar detects it as: a cyclist with probability 0.4; a pedestrian with probability 0.4; and a truck with probability 0.2.
● If the object is actually a truck, the lidar detects it as: a cyclist with probability 0.1; a pedestrian with probability 0.1; and a truck with probability 0.8.
If the camera system classifies the object as a cyclist, and the lidar system classifies it as a pedestrian, what are the probabilities (according to Bayes’ theorem) that the car should assign to the object being a cyclist, a pedestrian, and a truck? State any assumptions you make.
3 Hidden Markov Model Filtering (5 marks)
Let Xk e {1, 2, 3} for k > 0 be a three-state Markov chain with:
● Initial state pmf
π0 = -- --
and,
● State transition probability matrix
A = --
│ 0 0.2 1ǐ .
The states are observed through the measurement process Yk for k > 1 where p(Yk = ylXk = x) = N (y; x, 1).
If we receive the measurements Y1 = 1, Y2 = 2, Y3 = 3:
a) What is the most probable state at time k = 3 given the measurements?
b) What is the most probable (predicted) state at time k = 4 given the measurements?
Show all working.
Hint: The function N (y; µ, σ) is the Gaussian pdf evaluated at y with mean µ and variance σ 2 , defined as
N (y; µ, σ2 ) = exp ╱ - 、
2022-08-26