Some questions for Navigating Habitat Agent in Gazebo about ros_x_habitat HOT 1 OPEN

Hi @SepMJ if this's a Habitat agent navigating in a Gazebo-simulated environment then I wouldn't say the performance is unusual... the experiment we discussed in Section VI had fairly straight-forward goals for the agent, and it did take me quite a few tries before getting the agent reach the goal position. But that experiment was only for demoing that we can plug a Habitat agent into Gazeo, so we didn't fine tune the agent so it reaches optimal performance.

But if you do want to get it perform well in Gazebo, probably you'll have to train your own RL agent using Gazebo-simulated scenes instead of using the pre-trained agents. Those were trained on more photorealisitc Matterport3D/Gibson dataset and probably won't generalize well to a Gazebo environment.

from ros_x_habitat.