We document the HabiCrowd dataset as follows.
|- configs/ (Config for all baselines.)
|- baseline_<name>.yaml (Specified config on hyperparameters of each algorithm)
|- challenge_crowdnav.local.rgbd.yaml (Shared config about environment of all algorithms.)
|- crowd-nav
|- bot_config_v2.1/ (New version config for bots)
|- train (Train split)
|- <scene_name>.json
|- val (Val split)
|- <scene_name>.json
|- crowdnav_hm3d_v2.1/ (New version configs for PointNav and ObjectNav)
|- train/ (Train split)
|- content
|- <scene_name>.json.gz
|- train.json.gz
|- val/ (Val split)
|- content
|- <scene_name>.json.gz
|- val_mini/ (Demo split)
|- content
|- <scene_name>.json.gz
|- entities (Folder contains glb files of virtual humans)
|- textures
|- human_<i>.glb
|- human_<i>.mtl
|- human_<i>.object_config.json
As per 7th August 2023, we have included a total of 480 configurations of human dynamics to HabiCrowd. The new version for bot_config is 2.1.
As per 10th August 2023, we also included a new set of configs to handle with Instance Image Navigation. Instruction can be found in the main site.
For hyperparameters of baselines as well as simulator, see configs.
For virtual human parameters settings of each scene, see crowd-nav/bot_configs_v1.0. Each file specifies the number of virtual humans used, the initial position, rotation, and their desired linear, angular velocities.
For the navigation settings, see crowd-nav/crowdnav_hm3d. The folder is adapted from Objectnav_HM3D.
For the material of humans, see entities.
We will release weights of pre-trained model after finishing code instructions.
Note that, this is only the documentation of HabiCrowd. If you want to install as well as run code, we recommend to read the instructions from the main site.