Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

try out dreamer #62

Open
3 of 5 tasks
Tracked by #51
Armandpl opened this issue Jan 22, 2024 · 2 comments
Open
3 of 5 tasks
Tracked by #51

try out dreamer #62

Armandpl opened this issue Jan 22, 2024 · 2 comments

Comments

@Armandpl
Copy link
Owner

Armandpl commented Jan 22, 2024

  • make reset faster by having a pid bring the motor to a stop. then maybe have a second one stop the pendulum from oscillating faster
    • also maybe make the angle threshold a bit wider for the pendulum. I think sometime it is outside of the bounds? maybe the encoders are slipping? maybe the seeduino is missing some encoder step? maybe the logic level converter is causing issue? maye the cables are too long?
      • could print the angle when we hit the timeout
        • convert from rad to deg
  • setup dreamer in sim
  • setup dreamer on the robot
@Armandpl Armandpl mentioned this issue Jan 22, 2024
18 tasks
@Armandpl
Copy link
Owner Author

Armandpl commented Feb 8, 2024

dreamer is especially intersting here bc the dynamics of the system are probably trivial to learn + it's costly to collect steps on the real robot

@Armandpl
Copy link
Owner Author

in the same vibe we could learn a neural sim, maybe use a gpt model like comma is doing? check out karpathy's videos
having a good neural sim could be useful to tune/select good controllers, instead of hoping for rl agent convergence

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant