2024-01-26 and 2024-01-27 Meeting Agendas

[simojo]

:robot: JJ

• I have this laser driver that only works for Arduino code. How should I go about getting it to interface with ROS? I would rather not use an intermediate Arduino platform (for saving weight).
• Pausing the gazebo physics in conjunction with the PX4 program--I need to get it synchronized--any tips?
• Discussing timeline and getting strict due dates for everything.

[simojo]

• Understanding how to explain policy gradients and being able to define all of the variables.
  • I have been using OpenAI's Deep RL book as reference.

[simojo]

:atom: Willey

• Understanding inertial tensors better than I do already
• Standard way of explaining rotational matrices and why they're important

[simojo]

• Plates printed, trying to simulate this with much difficulty
• JJ did a gazebo rangefinder project almost 10 years ago. She's sure that things have changed. She wants to understand the connection between the simulation and what will be on the robot.
• A Q function is used in Q learning, and, it takes $s$ and $a$ and, using the Q table, you plug in all of the values, and, for all possible states and actions, you calculate values for all possible states and actions.
• JJ: forget about writing right now. I should be able to catch up later. As a collective group, they wanted to provide students with a good time frame. It is important to get the work done and to get good experimental data. In 2 weeks, if we're not making steady enough progress, we will need to come up with some kind of back tracking plan. If I tweak the parameters with gradient descent, it should work relatively well.
• JJ says that this may not be the best approach.
• Dedicate the rest of january and february to getting this work done.
• Doing the discrete approach will be easier to implement. If I am going to go for the combination approach, JJ says implement it fully discretely first, and then see how it does. Then, add on the continuous part!!!
• It may be possible to do a combination of discretized and continuous spaces.

[simojo]

• My project can still be successful even if I did not train successfully. I can still train and get failing results. I'm demonstrating what I tried and what didn't work. -- something to keep in mind.

[simojo]

My next steps

• Getting the simulation environment up and running with the sensors and 3D parts working.
• Simulate laser sensors
• explore learning strategies and discretizing things
• By mid-february, we can hope that the base line simulation is working
• Keep documenting all work and regularly updating repository.