[Python] How can I write a reward function for an aloha stationary robot setup that is...

My capstone project team and I are trying to create a reward function using python code in a Linux environment. We have been utilizing an aloha stationary robot setup connected to a computer running a Linux environment to input code on. Our robot setup is complete with sensors and cameras to help the robot learn and replicate tasks.

We already figured out how to run code where we can perform and record trials of specific tasks like locating and picking up a can and placing the can into a roll of tape. My team has recorded over 200 trials of performing this task, and now we need the robot setup to use those recorded trials to try and perform the task itself.

My team has already attempted to come up with a code where the robot can learn from the recorded trials and perform the task correctly, but we failed to get the robot to learn when it did something wrong and the robot never came close to replicating our recorded trials.

We need a solution for a python code that will allow the robot setup to look at the recorded trials, then try to accomplish the goals of locating the can, picking up the can, moving the can through the air, and setting the can into the roll of tape. We need a reward function to help our robot perform each of these steps correctly.

We already tried running code where the robot setup would look at the recorded trials and try to accomplish what it thought was trying to be done, but the reward function would reward the robot whether it correctly performed the task or not. We couldn't find a way to tell the robot arms if it did something incorrectly through the reward function.

Continue reading...