I saw in the FAQ that the evaluation accepts Discrete actions as well. I was wondering how to enable that. When I tried to set is_discrete: false in the file locobot_p2p_nav_house.yaml, it gave me the error “Locobot does not support discrete actions”
Thakns for your question.
As part of our collaboration with Facebook, the top five teams from the Habitat Challenge will automatically participate in the our Challenge in Real World phase. They are using discrete actions for navigation. That’s why in the FAQ, we mention that we also accept discrete actions from them. I apologize for the ambiguity of the wording.
Currently training in iGibson with Locobot only supports continuous action, as we believe the best action space for natural, smooth, and efficient navigation will be continuous velocities (linear and angular).
Are you planning to train with discrete actions? If you have very strong reasons, let us know and we can discuss further.
Thank you for you response and clarification. We just thought discrete actions would be simpler to create a plan out of. We will get back to you incase we need any modification.