ac-93 / tactile_gym Goto Github PK

download the model files and load to evaluation env...

pybullet build time: Nov 26 2020 23:09:05
Loaded EGL 1.5 after reload.
GL_VENDOR=NVIDIA Corporation
GL_RENDERER=GeForce RTX 2080/PCIe/SSE2
GL_VERSION=4.6.0 NVIDIA 460.80
GL_SHADING_LANGUAGE_VERSION=4.60 NVIDIA
Version = 4.6.0 NVIDIA 460.80
Vendor = NVIDIA Corporation
Renderer = GeForce RTX 2080/PCIe/SSE2
ven = NVIDIA Corporation
ven = NVIDIA Corporation
Traceback (most recent call last):
File "demo_trained_agents.py", line 48, in
final_evaluation(
File "/home/robot/anaconda3/envs/robel/lib/python3.8/site-packages/tactile_gym-0.0.1-py3.8.egg/tactile_gym/sb3_helpers/eval_agent_utils.py", line 112, in final_evaluation
model = sb3.PPO.load(model_path)
File "/home/robot/anaconda3/envs/robel/lib/python3.8/site-packages/stable_baselines3-1.1.0a11-py3.8.egg/stable_baselines3/common/base_class.py", line 650, in load
data, params, pytorch_variables = load_from_zip_file(path, device=device, custom_objects=custom_objects)
File "/home/robot/anaconda3/envs/robel/lib/python3.8/site-packages/stable_baselines3-1.1.0a11-py3.8.egg/stable_baselines3/common/save_util.py", line 402, in load_from_zip_file
data = json_to_data(json_data, custom_objects=custom_objects)
File "/home/robot/anaconda3/envs/robel/lib/python3.8/site-packages/stable_baselines3-1.1.0a11-py3.8.egg/stable_baselines3/common/save_util.py", line 164, in json_to_data
deserialized_object = cloudpickle.loads(base64_object)
AttributeError: Can't get attribute '_make_cell' on <module 'cloudpickle.cloudpickle' from '/home/robot/anaconda3/envs/robel/lib/python3.8/site-packages/cloudpickle/cloudpickle.py'>
Destroy EGL OpenGL window.

When I try to run train_agent.py, it shows
ImportError: cannot import name 'ImpalaCNN' from 'tactile_gym.sb3_helpers.custom.custom_torch_layers'

Hi authors, thank you very much for sharing this awesome repo!

I'm currently using GelSight Mini from GelSight Inc., and am wondering if I can somehow use the codebase to simulate deformations of the GelSight Mini sensor.

I think the GelSight Mini sensor is very similar to Digit, so I imagine this simulator can simulate it by changing some parameters, such as parameters that define sensor size, geometries of the gel, depth of the gel, pose of the sensor, etc. Even though the simulation won't be perfect, it can provide us with some reasonable features such as edges.

So, could you share some workarounds or tips to adopt different simulators?

Thanks!

@ac-93

I meet the error when to run script "python examples/demo_trained_agent.py -env='env_name' -obs='obs_type' -algo='algo_name'
". The models have downloaded and placed in the path of "tactile_gym/examples/enjoy".
Would you mind sharing the method to solve this problem?

I would like to inquire about how to modify the object_push trajectory or customize the environment. What should I do?

could you upload json param file?

It's really a very nice project, tactile robotics is a novel topic.

I am the developer of DI-engine, looking for some interesting environments to apply advanced DRL algorithms and collect meaningful demonstrations. Are you willing to develop more examples and benchmark results together, or any other ideas?

Loaded EGL 1.4 after reload.Unable to create EGL context(eglError:12292)Traceback(most recent call last):File "train agent.py", line 113,in train agent(File "train agent.py", line 41,in train agentenv = make training envs(env id, env args, rl params, save dir)File "/home/ubuntu1804/tactile_gym/tactile_ gym/sb3 helpers/rl utils.py", line15,in make training envs
env = make vec env
File "/home/ubuntu1804/anaconda3/envs/python38/lib/python3.8/site-packages/stable baselines3/common/env util.py", line 105, in make vec envreturn vec_env_cls([make env(i + start index)for i in range(n envs)], **vecenv kwargs)File "/home/ubuntu1804/anaconda3/envs/python38/lib/python3.8/site-packages/stable baselines3/common/vec _env/subproc_vec_env.py", line 111,in _initobservation space, action space = self.remotes[0].recv()File "/home/ubuntu1804/anaconda3/envs/python38/lib/python3.8/multiprocessing/c
onnection.py"，line 250，inrecvbuf = self. recv bytes()File "/home/ubuntu1804/anaconda3/envs/python38/lib/python3.8/multiprocessing/connection.py",line 414,in recv bytesbuf = self. recv(4)File "/home/ubuntu1804/anaconda3/envs/python38/lib/python3.8/multiprocessing/connection.py",line 379,in recv
chunk = read(handle,remaining)ConnectionResetError:[Errno 104]connection reset by peer

For code reason,I want to use tactile_gym in pybullet 3.2.5. So I change the requirements.txt of tactile_gym version to >=3.1.0. But some example tests(not all, demo_roll_env could be run) occured problems such as :

python examples/demo_example_arm_env.py
...
Traceback (most recent call last):
  File "examples/demo_example_arm_env.py", line 85, in <module>
    main()
  File "examples/demo_example_arm_env.py", line 45, in main
    env = ExampleArmEnv(
  File "/home/demons/anaconda3/envs/aerm/lib/python3.8/site-packages/tactile_gym-0.0.1-py3.8.egg/tactile_gym/rl_envs/example_envs/example_arm_env/example_arm_env.py", line 43, in __init__
    self.arm_type = env_modes["arm_type"]
KeyError: 'arm_type'
...

python examples/demo_edge_auto_env.py
...
MotionThreadFunc thread started
Traceback (most recent call last):
  File "examples/demo_edge_auto_env.py", line 122, in <module>
    main()
  File "examples/demo_edge_auto_env.py", line 47, in main
    env = EdgeFollowAutoEnv(
  File "/home/demons/anaconda3/envs/aerm/lib/python3.8/site-packages/tactile_gym-0.0.1-py3.8.egg/tactile_gym/rl_envs/exploration/edge_follow/edge_follow_auto_env.py", line 53, in __init__
    self.stimulus = env_modes["stimulus"]
KeyError: 'stimulus'
...

python examples/demo_push_env.py
...
b3Printf: /home/demons/anaconda3/envs/aerm/lib/python3.8/site-packages/tactile_gym-0.0.1-py3.8.egg/tactile_gym/assets/robot_assets/mg400/tactip/mg400_with_mini_right_angle_tactip.urdf:63: cannot find 'mg400/meshes/base_link.STL' in any directory in urdf path

Warning: b3Error[examples/Importers/ImportURDFDemo/BulletUrdfImporter.cpp,121]:

Warning: Could not parse visual element for Link:

...

It seems that different version of pybullets are not compatible well. Do you have some advice about version compatibility about bullet? Or how do I change the tactile_gym to make it work in upgraded pybullet.

The sb3_contrib package can not import the RAD_SAC and RAD_PPO,
And current sb3_contrib repo can only support TQC and QR-DQN.
So we can not reproduce the performance now?

I still encounter errors when following this procedure.https://github.com/ac-93/stable-baselines3-contrib/blob/RAD_aug_buffers/README.md