Build Apollo 2.5 with Perception Visualizer

build_apollo_2.5_visualizer.md

Requirement

• GPU: GTX-1080 or GTX-1080TI

• OS: Ubuntu 16.04.7 (https://releases.ubuntu.com/16.04/)

• Apollo 2.5: https://github.com/ApolloAuto/apollo/tree/r2.5.0

• Rosbag file: https://github.com/ApolloAuto/apollo/releases/download/v2.5.0/demo_2.5.bag. After downloading, move the file to apollo/data/bag

Step 1: Setup Environment

Host Machine:

• Install Nvidia driver:

   sudo add-apt-repository ppa:graphics-drivers/ppa
   sudo apt-get update
   sudo apt-get install nvidia-384
  

• Install Cuda-8.0:

  • Follow this instruction: https://rodrigodzf.github.io/setup/cuda/2019/04/15/cuda-setup.html

• Install Docker:

  • Follow this instruction: https://docs.docker.com/engine/install/ubuntu/

Docker

• Go to apollo folder. Open file ./docker/script/dev_start.sh, then adding two lines in command docker run -it

  -v /usr/src:/usr/src
  -v /lib/modules:/lib/modules
  

• Start and get into the docker with root authority

  ./docker/scripts/dev_start.sh
  docker exec -it apollo_dev /bin/bash
  

• If you receive the error 408 timeout when run dev_start.sh, please follow this instruction to set network mtu to 1200 (http://www.microhowto.info/howto/change_the_mtu_of_a_network_interface.html)

• Install nvidia-384 driver and CUDA-8.0: same as host machine.

• Install cuDNN (you may need to create an Nvidia developer account to download the install file, so I upload it to Teams https://teams.microsoft.com/_#/files/General?threadId=19%3A5f36071370c94195b1175facc74ff92a%40thread.tacv2&ctx=channel&context=cudnn7&rootfolder=%252Fsites%252FVinAI-AutoPilot%252FShared%2520Documents%252FGeneral%252FPlatformTeam%252FApollo%2520study%252FInstallation%2520file%252Fcudnn7). After download:

  sudo dpkg -i libcudnn7*.deb
  

• Install GLFW3.2: Please use the script in this folder https://teams.microsoft.com/_#/files/General?threadId=19%3A5f36071370c94195b1175facc74ff92a%40thread.tacv2&ctx=channel&context=Installation%2520file&rootfolder=%252Fsites%252FVinAI-AutoPilot%252FShared%2520Documents%252FGeneral%252FPlatformTeam%252FApollo%2520study%252FInstallation%2520file.

• Create config file in /usr/local/cuda/ with name nvblas.conf. Put this line into the file:

  NVBLAS_CPU_BLAS_LIB  /usr/lib/libopenblas.so 
  

• Run:

  export NVBLAS_CONFIG_FILE=/usr/local/cuda/nvblas.conf
  

• Commit a new docker image (in host):

  docker commit CONTAINER_ID apolloauto/apollo:NEW_DOCKER_IMAGE_TAG
  

  You can get container id by using the command below and find apollo_dev docker:

  docker container list
  

• Exit docker then start the new docker image (in host) and get into docker:

  ./docker/scripts/dev_start.sh -l -t NEW_DOCKER_IMAGE_TAG
  ./docker/scripts/dev_into.sh
  

Step 2: Build source code:

Modify and remove unused code:

• Go to file ./modules/perception/obstacle/onboard/visualization_subnode.cc. Remove those lines:

  if (lidar_event_id_ != -1 || fusion_event_id_ != -1) {
      lidar_object_data_ = dynamic_cast<LidarObjectData*>(
          shared_data_manager_->GetSharedData("LidarObjectData"));
      if (lidar_object_data_ == nullptr) {
        AERROR << "Failed to get LidarObjectData.";
        return false;
      }
      AINFO << "Init shared datas successfully, data: "
            << lidar_object_data_->name();
    }
  

  scene_shared_data_ = dynamic_cast<SceneSharedData*>(
       shared_data_manager_->GetSharedData("SceneSharedData"));
  
   if (scene_shared_data_ == nullptr) {
     AERROR << "Failed to get SceneSharedData.";
     return false;
   }
  

  SubnodeHelper::ProduceSharedDataKey(trigger_ts, trigger_device_id,
                                           &data_key_sensor);
  SetLidarContent(data_key_sensor, content, timestamp);
  

  AINFO << "set lidar content";
  SubnodeHelper::ProduceSharedDataKey(trigger_ts, trigger_device_id,
                                             &data_key_sensor);
  SetLidarContent(data_key_sensor, content, timestamp);
  

• Go to file ./modules/perception/onboard/event_manager.cc. Add those lines below line 94, above queue->pop(event);

  if(queue->size() == 0){
      return true;
  }
  

• Go to file ./modules/perception/obstacle/camera/visualizer/glfw_fusion_viewer.cc. Add this line above glfwCreateWindow called.

  glfwWindowHint(GLFW_CONTEXT_CREATION_API, GLFW_EGL_CONTEXT_API);
  

• Go to file ./module/perception/onboard/subnode.cc. Remove those lines:

  if (type_ == DAGConfig::SUBNODE_IN) {
      AINFO << "Subnode == SUBNODE_IN, EXIT THREAD. subnode:" << DebugString();
      return;
  }
  

Build & Run

• In Docker:

  ./apollo.sh build_opt_gpu
  

• After build, run bootstrap:

  bootstrap.sh
  

• Open another terminal tab, get into the docker by using ./docker/script/dev_into.sh. Run rosbag file.

  rosbag play data/bag/demo_2.5.bag -l --clock
  

• Keep rosbag playing, back to previous tab and launch perception with visulizer:

  ./scripts/perception_offline_visualizer.sh
  

• The popup below will display: