deep-activate-localization

Verified Setup On:

Operating System: Ubuntu 20.04.2 LTS
Nvidia-Driver Version: 460.73.01
CUDA Version: 11.2

Wiki Page:

For more details to train/test particle filter or rl agent refer.

Installation:

Option1 : Installation from source refer
Option2 : Installation from singularity image refer Note: modifying locobot urdf can be tricky in this case.

Useful References:

Install Cuda Without Root link

   sh cuda_10.1.105_418.39_linux.run --silent --toolkit --toolkitpath=$HOME/tkit

Install Cudnn in Ubuntu link

Steps to setup project:

Install required Nvidia Driver + CUDA + CUDNN for the system. Also refer igibson documentation for system requirements link.
Install virtual environment/package management platform like anaconda/miniconda or python virtualenv. Following assumes anaconda is installed.
Create conda environment

conda env create -f environment.yaml source activate igibson export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$CONDA_PREFIX/lib/ ``` 3. Setup iGibson2.0 environment (forked repository). For latest installation procedure always refer to official doc link

   <root_folder>$ git clone --branch master https://github.com/suresh-guttikonda/iGibson.git --recursive
   <root_folder>$ cd iGibson
   <root_folder>/iGibson$ pip3 install -e .

Test igibson installation is successful

   <root_folder>/iGibson$ python
   >>> import igibson

Download required igibson's assets (robot's urdf, demo apartment, and others). For more datasets refer

   <root_folder>/iGibson$ python -m igibson.utils.assets_utils --download_assets
   <root_folder>/iGibson$ python -m igibson.utils.assets_utils --download_demo_data

For locobot robot, we modify urdf file by adding additional lidar sensor 'scan_link' link and joint to 'base_link'. For reference, turtlebot's urdf file has the sensor by default.

   <root_folder>/iGibson$ vi igibson/data/assets/models/locobot/locobot.urdf
 
 <link name="scan_link">
   <inertial>
      <mass value="0"/>
      <origin xyz="0 0 0"/>
      <inertia ixx="0.0001" ixy="0.0" ixz="0.0" iyy="0.0001" iyz="0.0" izz="0.0001"/>
   </inertial>
   <visual>
      <geometry>
         <cylinder length="0.02" radius="0.01"/>
      </geometry>
      <origin rpy="0 0 0" xyz="0.001 0 0.05199"/>
   </visual>
 </link>
 <joint name="scan_joint" type="fixed">
   <parent link="base_link"/>
   <child link="scan_link"/>
   <origin rpy="0 0 0" xyz="0.10 0.0 0.46"/>
   <axis xyz="0 0 1"/>
 </joint>

8. Test tensorflow installation is successful

   <root_folder>$ python
   >>> import tensorflow as tf
   >>> tf.config.list_logical_devices()

Install TF Agents (forked repository)

   <root_folder>/iGibson$ cd ..
   <root_folder>$ git clone --branch cvpr21_challenge_tf2.4 https://github.com/suresh-guttikonda/agents/ --recursive
   <root_folder>$ cd agents
   <root_folder>/agents$ pip3 install -e .

Get the Active Localization code + pretrained checkpoints

   <root_folder>/agents$ cd ..
   <root_folder>$ git clone https://github.com/suresh-guttikonda/deep-activate-localization.git --recursive

Test pretrained particle filter + igibson environment with obstacle avoidance agent, result will be stored to test_output directory.

   <root_folder>$ cd deep-activate-localization/src/rl_agents
   <root_folder>/deep-activate-localization/src/rl_agents$ python -u test_pfnet.py \
      --pfnet_loadpath=./pfnetwork/checkpoints/pfnet_igibson_data/report/rs_rgbd/checkpoint_28_0.065/pfnet_checkpoint \
      --obs_mode=rgb-depth \
      --custom_output rgb_obs depth_obs likelihood_map obstacle_obs \
      --scene_id=Rs \
      --num_eval_episodes=1 \
      --agent=avoid_agent \
      --init_particles_distr=uniform \
      --init_particles_std 0.2 0.523599 \
      --particles_range=100 \
      --num_particles=250 \
      --transition_std 0.04 0.0872665 \
      --resample=true \
      --alpha_resample_ratio=0.95 \
      --global_map_size 100 100 1 \
      --window_scaler=1.0 \
      --config_file=./configs/locobot_pfnet_nav.yaml \
      --device_idx=0 \
      --seed=15

Test pretrained SAC agent for igibson environment as follows (for global localization).

   <root_folder>$ cd deep-activate-localization/src/rl_agents
   <root_folder>/deep-activate-localization/src/rl_agents$ python -u test_rl_agent.py \
      --root_dir=checkpoints/train_rl_uniform_2.0_box_50 \
      --num_eval_episodes=10 \
      --use_tf_functions=False \
      --agent=sac_agent \
      --eval_deterministic=True \
      --is_localize_env=True \
      --config_file=./configs/locobot_point_nav.yaml \
      --gpu_num=0 \
      --init_env_pfnet=True \
      --init_particles_distr=uniform \
      --init_particles_std=0.2,0.523599 \
      --particles_range=200 \
      --num_particles=1000 \
      --resample=True \
      --alpha_resample_ratio=0.99 \
      --transition_std=0.04,0.0872665 \
      --obs_mode=rgb-depth \
      --custom_output=rgb_obs,depth_obs,likelihood_map \
      --num_clusters=10 \
      --global_map_size=100,100,1 \
      --window_scaler=1.0 \
      --pfnet_load=./pfnetwork/checkpoints/pfnet_igibson_data/report/rs_rgbd/checkpoint_28_0.065/pfnet_checkpoint \
      --use_plot=True \
      --store_plot=True \
      --seed=1198

Steps to install singularity:

Enable NeuroDebian repository by following instructions on neuro-debian. Also refer FAQ.

Install Singularity-Container as follows:

$ sudo apt-get install -y singularity-container

For more details, refer: singularity guide

Steps to create singularity image to setup project:

We assume corresponding nvidia-driver is already installed in host machine.

Bootstrap container definition file. 'sudo' ensure proper privileges get assigned.

# command takes few minutes to completely build image
<path_to_def_file>$ sudo singularity build tensorflow_latest-gpu.sif tensorflow_latest-gpu.def

After successful build, verify that packages such as tensorflow+gpu are installed correctly.
Note: --nv ensure cuda related packages are loaded

<path_to_sif_file>$ sudo singularity shell --nv tensorflow_latest-gpu.sif

# get driver details if above assumptions are satisfied
~> nvidia-smi

# activate virtualenv
~> source /opt/venvs/py3-igibson/bin/activate

~> python
>>> import tensorflow as tf
>>> tf.__version__
>>>> tf.config.list_physical_devices() => returns list of available devices

Run demo igibson gui example. Following should open window with robot otherwise we see and error 'ERROR: Unable to initialize EGL'

# Note: don't run with sudo once image is created
<path_to_def_file>$ singularity shell --nv -B /usr/share/glvnd tensorflow_latest-gpu.sif

~> source /opt/venvs/py3-igibson/bin/activate

# launch igibson gui example
(py3-igibson)~> python -m gibson2.examples.demo.env_example

To Bind/Mount host machine directories within container use --bind ref

<path_to_sif_file>$ singularity shell --nv --bind /usr/share/glvnd,./src:/mnt/src tensorflow_latest-gpu.sif

dhonerkamp / deep-activate-localization Goto Github PK