{"id":362,"date":"2020-03-26T19:15:10","date_gmt":"2020-03-26T23:15:10","guid":{"rendered":"http:\/\/mrsdprojects.ri.cmu.edu\/2020teamb\/?page_id=362"},"modified":"2021-09-08T18:26:53","modified_gmt":"2021-09-08T22:26:53","slug":"dynamics-and-controls-subsystem","status":"publish","type":"page","link":"https:\/\/mrsdprojects.ri.cmu.edu\/2020teamb\/implementation\/dynamics-and-controls-subsystem\/","title":{"rendered":"Dynamics and Controls Subsystem"},"content":{"rendered":"

Dynamics and Controls Subsystem Progress<\/h1>\n

The Dynamics and Control subsystem is where vehicle simulator development, vehicle dynamics and tire modeling, online model learning, and vehicle controls are designed and implemented. The initial vehicle model that has been implemented as a baseline is a dynamic bicycle model with parameters that map to our test vehicle platform. The next iteration would be to implement a neural network or model based reinforcement learning technique for representing the vehicle dynamics by taking in state feedback from simulation or on-board sensors and training the model to learn edge weights that allow accurate predictions of the vehicles future states.<\/p>\n

January 6th, 2020<\/h3>\n

Initialized BitBucket repository that will be used for revision control and documentation of ongoing software development within this subsystem.<\/p>\n

January 24th, 2020<\/h3>\n

After doing immense amounts of research on different environments that could be used for performing vehicle simulation, our team decided to move forward with developing our simulation within MATLAB\/Simulink. This had to do with the fact that compared to CARLA or Gazebo, we had more freedom for design choices, and had strong documentation and resources available from MathWorks, whereas the other two mentioned environments would limit our ability to model the nonlinear dynamics of our vehicle.<\/p>\n

February 7th, 2020<\/h3>\n

A pipeline diagram that lays out the overall structure and flow of data transfer throughout the simulator and subsequent robot was created, which we believe will be useful to refer to when constructing our software. This allows us to make more informed decisions about inputs and outputs from various functions, and helps determine what data and methods should be encased in class objects.<\/p>\n

\"\"<\/p>\n

 <\/p>\n

February 21st, 2020<\/h3>\n

First iteration of vehicle simulation was successfully implemented within MATLAB, with the first demonstration consisting of a simple PI longitudinal controller. A visualization of this can be seen below, where the blue rectangle represents the vehicle and the dotted red line represents the straight path:<\/p>\n

\"\"<\/p>\n

The visualization script uses homogeneous transformations to plot how the vehicle travels through space.<\/p>\n

March 6th, 2020<\/h3>\n

We have been prototyping various control algorithms for the lateral control of our robotic system. We are looking at methods that we can start off by implementing simplified versions, and then add in more complex features over time. We will be starting off with linearized tire dynamics, and fixed parameter vehicle model, and moving forward we will be implementing a fiala brush tire model based on data gathered during testing, along with implementing a learning-based approach to update model parameters on the vehicle as the system traverses a trajectory.<\/p>\n

March 24th, 2020<\/h3>\n

So far for lateral control we have been prototyping a Stanley controller, which has provided stable control, but has also produced a fairly large cross-track error, as can be seen in the image below:<\/p>\n

\"\"<\/p>\n

 <\/p>\n

This method is currently being refined.<\/p>\n

<\/h3>\n

March 28th, 2020<\/strong><\/h3>\n

Today, we successfully produced a prototyping circuit for the low-level vehicle controls. We are currently using a PWM board from Adafruit, that we communicate with over I2C from our Jetson Xavier, to send control signals to our vehicle’s electronic speed controller (ESC). With this circuit, we have been able to adjust the steering angle and the drive motor speed. Going forward, this will be used as a way to prototype onboard vehicle controls, once they are transferred from simulation.<\/p>\n

\"\"<\/p>\n

 <\/p>\n

April 1st, 2020<\/h3>\n

Today, the a simplified version of the Stanley controller that has been worked on within simulation has been tuned and is now providing considerably better tracking performance. A test trajectory and the vehicle’s subsequent tracking performance can be seen below:<\/p>\n