运行neural-active-slam问题记录

先建立了一个slam文件夹，后期把habitat-sim，habitat-api，nerual-active-slam都放在这个文件夹下。
1.先安装habitat-sim:

git clone https://github.com/facebookresearch/habitat-sim.git
cd habitat-sim; git checkout 9575dcd45fe6f55d2a44043833af08972a7895a9; 
pip install -r requirements.txt; 
python setup.py install --headless

运行到python setup.py install --headless时候出错，我的是有显示器 ，所以我把最后的headless去掉运行。错误是：

···
raise RuntimeError(
RuntimeError: CMake must be installed to build the following extensions: habitat_sim._ext.habitat_sim_bindings

根据查询google，在网页https://github.com/Greenwolf/social_mapper/issues/25中找到

I ran into this issue as well. I am using OS X 10.13 and have a Python virtual environment that I am installing the requirements to.

I ran `pip install cmake` and re-ran `pip install -r requirements.txt` and I no longer received the error.

按照此人方法，运行了pip install cmake确实出现了不一样的东西如下

Collecting cmake
  Downloading cmake-3.18.2.post1-py3-none-manylinux1_x86_64.whl (18.2 MB)
     |████████████████████████████████| 18.2 MB 1.4 MB/s 
Installing collected packages: cmake
Successfully installed cmake-3.18.2.post1

接着运行pip install -r requirement.txt，这下运行了很长时间，最后得到：

···
Stored in directory: /tmp/pip-ephem-wheel-cache-1ughjamc/wheels/35/76/38/fff4317b0c3c1d5375c4f95908dacfb40495de8fc37421c636
Successfully built magnum
Installing collected packages: magnum
Successfully installed magnum-0.0.0

应该算是成功了安装构建了habitat-sim。
2.接下来是habitat-api，根据步骤安装没错误

git clone https://github.com/facebookresearch/habitat-api.git
cd habitat-api; git checkout b5f2b00a25627ecb52b43b13ea96b05998d9a121; 
pip install -e .

3.接着安装neural-slam，没问题

git clone --recurse-submodules https://github.com/devendrachaplot/Neural-SLAM
cd Neural-SLAM;
pip install -r requirements.txt

还要把数据集放进去，我下载了“gibson_habitat.zip”数据集，解压后放进/Neural-SLAM/data/scene_datasets/gibson中。下载了“pointnav_gibson_v1”改名为“v1”，放入/Neural-SLAM/data/datasets/pointnav/gibson/v1中。
接着验证依赖和数据都设置好了，运行：

python main.py -n1 --auto_gpu_config 0 --split va

出错如下：

Dumping at ./tmp//models/exp1/
Namespace(alpha=0.99, auto_gpu_config=0, camera_height=1.25, clip_param=0.2, collision_threshold=0.2, cuda=True, du_scale=2, dump_location='./tmp/', entropy_coef=0.001, env_frame_height=256, env_frame_width=256, eps=1e-05, eval=0, exp_loss_coeff=1.0, exp_name='exp1', frame_height=128, frame_width=128, gamma=0.99, global_downscaling=2, global_hidden_size=256, global_lr=2.5e-05, goals_size=2, hfov=90.0, load_global='0', load_local='0', load_slam='0', local_hidden_size=512, local_optimizer='adam,lr=0.0001', local_policy_update_freq=5, log_interval=10, map_pred_threshold=0.5, map_resolution=5, map_size_cm=2400, max_episode_length=1000, max_grad_norm=0.5, no_cuda=False, noise_level=1.0, noisy_actions=1, noisy_odometry=1, num_episodes=1000000, num_global_steps=40, num_local_steps=25, num_mini_batch=0, num_processes=1, num_processes_on_first_gpu=0, num_processes_per_gpu=11, obs_threshold=1, obstacle_boundary=5, pose_loss_coeff=10000.0, ppo_epoch=4, pretrained_resnet=1, print_images=0, proj_loss_coeff=1.0, randomize_env_every=1000, save_interval=1, save_periodic=500000, save_trajectory_data='0', seed=1, short_goal_dist=1, sim_gpu_id=0, slam_batch_size=72, slam_iterations=10, slam_memory_size=500000, slam_optimizer='adam,lr=0.0001', split='val', task_config='tasks/pointnav_gibson.yaml', tau=0.95, total_num_scenes='auto', train_global=1, train_local=1, train_slam=1, use_deterministic_local=0, use_gae=False, use_pose_estimation=2, use_recurrent_global=0, use_recurrent_local=1, value_loss_coef=0.5, vis_type=1, vision_range=64, visualize=0)
Loading data/scene_datasets/gibson/Cantwell.glb
2020-10-13 11:22:19,837 initializing sim Sim-v0
Renderer: GeForce GTX 1650/PCIe/SSE2 by NVIDIA Corporation
OpenGL version: 4.6.0 NVIDIA 440.95.01
Using optional features:
    GL_ARB_ES2_compatibility
    GL_ARB_direct_state_access
    GL_ARB_get_texture_sub_image
    GL_ARB_invalidate_subdata
    GL_ARB_multi_bind
    GL_ARB_robustness
    GL_ARB_separate_shader_objects
    GL_ARB_texture_filter_anisotropic
    GL_ARB_texture_storage
    GL_ARB_texture_storage_multisample
    GL_ARB_vertex_array_object
    GL_KHR_debug
Using driver workarounds:
    no-layout-qualifiers-on-old-glsl
    nv-zero-context-profile-mask
    nv-implementation-color-read-format-dsa-broken
    nv-cubemap-inconsistent-compressed-image-size
    nv-cubemap-broken-full-compressed-image-query
    nv-compressed-block-size-in-bits
I1013 11:22:22.971073 24180 simulator.py:80] Loaded navmesh data/scene_datasets/gibson/Cantwell.navmesh
2020-10-13 11:22:22,988 initializing task Nav-v0
2020-10-13 11:22:23,013 Computing map for data/scene_datasets/gibson/Cantwell.glb
/home/thomastao/anaconda3/lib/python3.8/site-packages/torch/nn/functional.py:3447: UserWarning: Default grid_sample and affine_grid behavior has changed to align_corners=False since 1.3.0. Please specify align_corners=True if the old behavior is desired. See the documentation of grid_sample for details.
  warnings.warn("Default grid_sample and affine_grid behavior has changed "
/home/thomastao/anaconda3/lib/python3.8/site-packages/torch/nn/functional.py:3384: UserWarning: Default grid_sample and affine_grid behavior has changed to align_corners=False since 1.3.0. Please specify align_corners=True if the old behavior is desired. See the documentation of grid_sample for details.
  warnings.warn("Default grid_sample and affine_grid behavior has changed "
Downloading: "https://download.pytorch.org/models/resnet18-5c106cde.pth" to /home/thomastao/.cache/torch/hub/checkpoints/resnet18-5c106cde.pth
100%|█████████████████████████████████████| 44.7M/44.7M [00:16<00:00, 2.76MB/s]
/home/thomastao/anaconda3/lib/python3.8/site-packages/torch/nn/functional.py:3447: UserWarning: Default grid_sample and affine_grid behavior has changed to align_corners=False since 1.3.0. Please specify align_corners=True if the old behavior is desired. See the documentation of grid_sample for details.
  warnings.warn("Default grid_sample and affine_grid behavior has changed "
/home/thomastao/anaconda3/lib/python3.8/site-packages/torch/nn/functional.py:3384: UserWarning: Default grid_sample and affine_grid behavior has changed to align_corners=False since 1.3.0. Please specify align_corners=True if the old behavior is desired. See the documentation of grid_sample for details.
  warnings.warn("Default grid_sample and affine_grid behavior has changed "
Time: 00d 00h 00m 00s, num timesteps 0, FPS 0,
	Rewards:
	Losses:
Time: 00d 00h 00m 00s, num timesteps 10, FPS 12,
	Rewards:
	Losses: Local Loss: 5.276,
Time: 00d 00h 00m 03s, num timesteps 20, FPS 5,
	Rewards:
	Losses: Local Loss: 5.327,
Time: 00d 00h 00m 08s, num timesteps 30, FPS 3,
	Rewards:
	Losses: Local Loss: 5.356,
Time: 00d 00h 00m 12s, num timesteps 40, FPS 3,
	Rewards:
	Losses: Local Loss: 5.150,
Time: 00d 00h 00m 16s, num timesteps 50, FPS 2,
	Rewards:
	Losses: Local Loss: 5.238,
Time: 00d 00h 00m 21s, num timesteps 60, FPS 2,
	Rewards:
	Losses: Local Loss: 5.395,
Time: 00d 00h 00m 25s, num timesteps 70, FPS 2,
	Rewards:
	Losses: Local Loss: 5.354,
Traceback (most recent call last):
  File "main.py", line 769, in <module>
    main()
  File "main.py", line 556, in main
    nslam_module(b_obs_last, b_obs, b_poses,
  File "/home/thomastao/anaconda3/lib/python3.8/site-packages/torch/nn/modules/module.py", line 722, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/home/thomastao/slam/Neural-SLAM/model.py", line 181, in forward
    resnet_output = self.resnet_l5(obs_last[:, :3, :, :])
  File "/home/thomastao/anaconda3/lib/python3.8/site-packages/torch/nn/modules/module.py", line 722, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/home/thomastao/anaconda3/lib/python3.8/site-packages/torch/nn/modules/container.py", line 117, in forward
    input = module(input)
  File "/home/thomastao/anaconda3/lib/python3.8/site-packages/torch/nn/modules/module.py", line 722, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/home/thomastao/anaconda3/lib/python3.8/site-packages/torch/nn/modules/conv.py", line 419, in forward
    return self._conv_forward(input, self.weight)
  File "/home/thomastao/anaconda3/lib/python3.8/site-packages/torch/nn/modules/conv.py", line 415, in _conv_forward
    return F.conv2d(input, weight, self.bias, self.stride,
RuntimeError: CUDA out of memory. Tried to allocate 72.00 MiB (GPU 0; 3.81 GiB total capacity; 938.31 MiB already allocated; 66.12 MiB free; 966.00 MiB reserved in total by PyTorch)
Exception ignored in: <function VectorEnv.__del__ at 0x7fa9518a8940>
Traceback (most recent call last):
  File "/home/thomastao/slam/Neural-SLAM/env/habitat/habitat_api/habitat/core/vector_env.py", line 487, in __del__
    self.close()
  File "/home/thomastao/slam/Neural-SLAM/env/habitat/habitat_api/habitat/core/vector_env.py", line 351, in close
    write_fn((CLOSE_COMMAND, None))
  File "/home/thomastao/anaconda3/lib/python3.8/multiprocessing/connection.py", line 206, in send
    self._send_bytes(_ForkingPickler.dumps(obj))
  File "/home/thomastao/anaconda3/lib/python3.8/multiprocessing/connection.py", line 411, in _send_bytes
    self._send(header + buf)
  File "/home/thomastao/anaconda3/lib/python3.8/multiprocessing/connection.py", line 368, in _send
    n = write(self._handle, buf)
BrokenPipeError: [Errno 32] Broken pipe

接着运行训练模型的命令：出错如下：

Traceback (most recent call last):
  File "main.py", line 33, in <module>
    args = get_args()
  File "/home/thomastao/slam/Neural-SLAM/arguments.py", line 197, in get_args
    assert torch.cuda.get_device_properties(i).total_memory \
AssertionError: Insufficient GPU memory

又回过头来运行验证的命令：

python main.py -n1 --auto_gpu_config 0 --split val

又出错如下：错误变了

Traceback (most recent call last):
  File "main.py", line 15, in <module>
    from env import make_vec_envs
  File "/home/thomastao/slam/Neural-SLAM/env/__init__.py", line 3, in <module>
    from .habitat import construct_envs
  File "/home/thomastao/slam/Neural-SLAM/env/habitat/__init__.py", line 9, in <module>
    from .exploration_env import Exploration_Env
  File "/home/thomastao/slam/Neural-SLAM/env/habitat/exploration_env.py", line 28, in <module>
    from env.habitat.utils.noisy_actions import CustomActionSpaceConfiguration
  File "/home/thomastao/slam/Neural-SLAM/env/habitat/utils/noisy_actions.py", line 9, in <module>
    from habitat.sims.habitat_simulator.action_spaces import (
ModuleNotFoundError: No module named 'habitat.sims.habitat_simulator.action_spaces'

截止到这里，我最困惑的一点是neural-slam和habitat-api和habitat-sim三个文件夹应该怎么放置。退而求其次，现在连habitat-api和habitat-sim这两个文件夹的放置都没搞明白。需要好好思考一下。