code for paper 'Learning View and Target Invariant Visual Servoing for Navigation'
Setup GibsonEnv
Download and install gibson environment from https://github.com/StanfordVL/GibsonEnv.
git checkout d3aa0a1
Generate Training and Testing data
python sample_image_pairs_with_common_area.py --scene_idx=0
python visualize_sampled_image_pairs.py
Learned Visual Servoing Model
To train a learned-vs model through DQN,
python vs_controller/train_DQN_vs_overlap.py
To evaluate the performance of the trained model,
python vs_controller/evaluate_DQN_vs.py
Classical VS Model
Test the vs model using sift and ground-truth depth,
python visual_servoing/test_IBVS_SIFT_interMatrix_gtDepth_Vz_OmegaY.py --scene_idx=$i
Generate Occupancy Map
Used some code from https://github.com/tenther/cs685-project.
To generate nice occupancy maps for the used Gibson environments,
sh rrt/run_make_rrt.sh