Robotic control problems are said to require continuous action spaces. But is that really necessary for granular control with real world robotics? At the end of the day, even human movement is restricted on some molecular level that does not correspond to truly 'continuous' / infinitely granular actions.

  • $\begingroup$ If you take an example of rotational robotic joint, let's say allowed rotational range is all 360 degrees. If you want to have precision of 1 degree you would need to have 360 separate actions, and you would need to calculate values of all those 360 actions and pick the best one. It's not computationally feasible to do that for large number of discrete actions $\endgroup$ – Brale_ Feb 11 at 11:16
  • $\begingroup$ @Brale_ RL generally handles that better than continuous action space. $\endgroup$ – user3180 Feb 12 at 5:39

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Browse other questions tagged or ask your own question.