IndexError: tensors used as indices must be long, int, byte or bool tensors

[logic110]

Hi, thank you for your great job about parkour. I encountered some problems when I ran "barrier_track.py". "IndexError: tensors used as indices must be long, int, byte or bool tensors." What I need to do?

[ZiwenZhuang]

Hi,
Could you explain more details about your problem?
The file barrier_track.py is not designed to be run alone.
As I remembered, "IndexError: tensors used as indices must be long, int, byte or bool tensors." happened when some of the field in the configuration becomes a float rather than an integer. But I need more details to help you.

Could you show more about how you run the script and what does the error message says?

Best,

[logic110]

Thank you for your fast reply! These are the wrong details.
python play.py --task a1_crawl --load_run /home/robot/parkour/parkour/legged_gym/legged_gym/field_a1/crawl_raw_919_ok
Importing module 'gym_38' (/home/robot/parkour/IsaacGym_Preview_4_Package/isaacgym/python/isaacgym/_bindings/linux-x86_64/gym_38.so)
Setting GYM_USD_PLUG_INFO_PATH to /home/robot/parkour/IsaacGym_Preview_4_Package/isaacgym/python/isaacgym/_bindings/linux-x86_64/usd/plugInfo.json
PyTorch version 2.1.0+cu121
Device count 1
/home/robot/parkour/IsaacGym_Preview_4_Package/isaacgym/python/isaacgym/_bindings/src/gymtorch
Using /home/.cache/torch_extensions/py38_cu121 as PyTorch extensions root...
Emitting ninja build file /home/.cache/torch_extensions/py38_cu121/gymtorch/build.ninja...
Building extension module gymtorch...
Allowing ninja to set a default number of workers... (overridable by setting the environment variable MAX_JOBS=N)
ninja: no work to do.
Loading extension module gymtorch...
Setting seed: 1
Using LeggedRobotField.init, num_obs and num_privileged_obs will be computed instead of assigned.
Not connected to PVD
+++ Using GPU PhysX
Physics Engine: PhysX
Physics Device: cuda:0
GPU Pipeline: enabled
/home/anaconda3/envs/parkour/lib/python3.8/site-packages/torch/functional.py:504: UserWarning: torch.meshgrid: in an upcoming release, it will be required to pass the indexing argument. (Triggered internally at ../aten/src/ATen/native/TensorShape.cpp:3526.)
return _VF.meshgrid(tensors, **kwargs) # type: ignore[attr-defined]
Total number of volume estimation points for each robot is: 2909
Traceback (most recent call last):
File "play.py", line 340, in
play(args)
File "/home/anaconda3/envs/parkour/lib/python3.8/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "play.py", line 165, in play
env.reset()
File "/home/robot/parkour/parkour/legged_gym/legged_gym/envs/base/base_task.py", line 114, in reset
obs, privileged_obs, _, _, _ = self.step(torch.zeros(self.num_envs, self.num_actions, device=self.device, requires_grad=False))
File "/home/robot/parkour/parkour/legged_gym/legged_gym/envs/base/legged_robot.py", line 97, in step
self.post_physics_step()
File "/home/robot/parkour/parkour/legged_gym/legged_gym/envs/base/legged_robot_field.py", line 121, in post_physics_step
return super().post_physics_step()
File "/home/robot/parkour/parkour/legged_gym/legged_gym/envs/base/legged_robot.py", line 135, in post_physics_step
self.check_termination()
File "/home/robot/parkour/parkour/legged_gym/legged_gym/envs/base/legged_robot_field.py", line 135, in check_termination
stepping_obstacle_info = self.terrain.get_stepping_obstacle_info(self.volume_sample_points.view(-1, 3))
File "/home/robot/parkour/parkour/legged_gym/legged_gym/utils/terrain/barrier_track.py", line 720, in get_stepping_obstacle_info
obstacle_info = self.track_info_map[
IndexError: tensors used as indices must be long, int, byte or bool tensors

[ZiwenZhuang]

What is your pytorch version? Probably because line 698 in barrier_track.py. Maybe try forcing the data type on every possible values?

[jdluuu]

I meet this question too. It can be fixed by change your pytorch to version 1.10.0

[jdluuu]

Besides, mismatched versions of pytorch and cuda worked fine on my computer (pytorch 1.10.0 + cuda 11.7, RTX 3090)

I meet this question too. It can be fixed by change your pytorch to version 1.10.0

[Shifters1]

I got the same error, using pytorch 1.10.0 + cuda 11.3, RTX 4090, is there any fix?

[guyo-shifters]

i fixed it by adding .long() to the indices in this line, then it works with newer versions of pytorch

[sandorfelber]

i fixed it by adding .long() to the indices in this line, then it works with newer versions of pytorch

Thanks! This worked for me too.

[CrazyWan528]

i fixed it by adding .long() to the indices in this line, then it works with newer versions of pytorch

Are you using a 4090 GPU? I tried pytorch2.1.0+cuda12.1, pytorch2.0.0+cuda11.8, and pytorch1.10.0+cuda11.3, and they all reported the same error! I'm not sure if my change to "torch.zeros_like(track_idx[0]).long()" on line 698 is correct or not.

[CrazyWan528]

i fixed it by adding .long() to the indices in this line, then it works with newer versions of pytorch

Based on your suggestion, I changed the .to(int) in line 737 to .long(), and added .long() after line 742 as well, and it then worked fine with the env of pytorch2.0.0+cuda11.8 on RTX4090.

[AlorithmKing]

change newer version of pytorch pytorch2.0.0+cuda11.8 and add .long()