search

devendrachaplot / Neural-SLAM

Pytorch code for ICLR-20 Paper "Learning to Explore using Active Neural SLAM"
http://www.cs.cmu.edu/~dchaplot/projects/neural-slam.html
MIT License
757 stars 143 forks source link

Separate the local policy module and test on the ground-truth map #24

Closed trqminh closed 3 years ago

trqminh commented 3 years ago

As I understand your paper and code, I think with the Exploration_Environment, at each time step, I can get the observation (the image from simulation), the global ground-truth map 480x480 pixels), and the local pose of the robot. Thereby, I can you this information and provide a goal (a pixel on the map) to feed to the local policy to navigate.

I think I'll modify the exploration_env.py so that I can return self.explorable_map and self.curr_loc_gt at each time step.

Is it possible for me to doing so? Thank you for your consideration.

devendrachaplot commented 3 years ago

Yes, that's correct, you can pass the ground truth map and pose as a part of the info dictionary from the Environment class in the exploration_env.py file.