Embedded Files
Project Overview
Project Overview
The objective of this project was to create a voice-controlled car that uses steady-state feedback to drive forward and turn.
The objective of this project was to create a voice-controlled car that uses steady-state feedback to drive forward and turn.
Steady State Scheme:
Steady State Scheme:
The model equations are set up to track the state of the car, given by optical encoders.
The model equations are set up to track the state of the car, given by optical encoders.
Closed-Loop Model Derivation:
Closed-Loop Model Derivation:
In an open-loop model, the input at each time step is a function of the determined operating velocity (calculated with least squares regression) and the physical behavior of each wheel. However, we can adjust the input at each state based on the state of each wheel. By doing this, we can adjust the input to make the car drive straight.
In an open-loop model, the input at each time step is a function of the determined operating velocity (calculated with least squares regression) and the physical behavior of each wheel. However, we can adjust the input at each state based on the state of each wheel. By doing this, we can adjust the input to make the car drive straight.
Afterward, the system eigenvalue can be determined by picking the weighting factors kL and kR. As this is a discrete system, we can choose our eigenvalues carefully to have stability without oscillations.
Afterward, the system eigenvalue can be determined by picking the weighting factors kL and kR. As this is a discrete system, we can choose our eigenvalues carefully to have stability without oscillations.
• n: The current timestep of the model. It is used to evaluate the model parameters at each time step.• d[n]: The total number of ticks advanced by either the right or left encoder since the initial time step. It is used to measure the distance traveled of the car.• v[n]: The discrete-time velocity (in units of ticks/timestep) of the wheel. It is used to measure the instantaneous speed of the car at a certain time step.• u[n]: The input voltage to the system delivered via PWM.• θ: The response to change in velocity in response to the input voltage. It is used to determine the expected change in velocity in response to a change in input.• β - A constant offset in velocity (units are ticks/timestep). β is dependent upon many physical factors, so it is determined empirically.
Turning:
Turning:
In our model, when the right wheel has moved further than the left, resulting in a negative delta term and causing the car to turn left, the control scheme compensates for this and turns the car back to the right. Therefore, we derived an equation for what nonzero term we would want our delta term to be given the distance between the center of the wheels (l), the operating point (v*, noted as v in the derivation), the time step (k), and the turning radius (r). This derivation comes from analyzing a right turn, meaning that we can just set delta to be negative for a left turn. When we want to go straight, we would simply set the delta term to be zero.
In our model, when the right wheel has moved further than the left, resulting in a negative delta term and causing the car to turn left, the control scheme compensates for this and turns the car back to the right. Therefore, we derived an equation for what nonzero term we would want our delta term to be given the distance between the center of the wheels (l), the operating point (v*, noted as v in the derivation), the time step (k), and the turning radius (r). This derivation comes from analyzing a right turn, meaning that we can just set delta to be negative for a left turn. When we want to go straight, we would simply set the delta term to be zero.
Adding Voice Commands:
Adding Voice Commands:
Using a microphone circuit with a low-pass filter to reject higher noise frequencies above the human range, we recorded many samples of our own voice commands. By using SVD and PCA on our data, we are able to separate clusters data points when projected onto the principal components. Finally, we can project a new voice command input onto the principal components and calculate the distance to each centroid of a cluster. If it is within a tolerance range and is over the acceptable volume threshold, it is then accepted as a command.
Using a microphone circuit with a low-pass filter to reject higher noise frequencies above the human range, we recorded many samples of our own voice commands. By using SVD and PCA on our data, we are able to separate clusters data points when projected onto the principal components. Finally, we can project a new voice command input onto the principal components and calculate the distance to each centroid of a cluster. If it is within a tolerance range and is over the acceptable volume threshold, it is then accepted as a command.
"Salmon"
"Salmon"
Drive Far
Drive Far
Salmon.mov"Hasagi"
"Hasagi"
Turn Left
Turn Left
Hasagi.MOV"Kote"
"Kote"
Drive Close
Drive Close
Kote.mov"Iron Man"
"Iron Man"
Turn Right
Turn Right
Iron Man.MOV
Page updated
Google Sites
Report abuse