RuntimeError when trying to run IV_SAC model

[AdityaGudimella]

Hello. Thank you for your great paper. I cloned the repo, and installed the requirements and am trying to run the IV_SAC experiment with Hopper. I don't have a mujoco license, so I replaced the mujoco env with PyBullet env. When I run python main.py --env HopperBulletEnv-v0 --model IV_SAC --net_seed 0 --env_seed 0 I get the error:

Variable._execution_engine.run_backward(
RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.FloatTensor [256, 1]], which is output 0 of AsStridedBackward0, is at version 2; expected version 1 instead. Hint: enable anomaly detection to find the operation that failed to compute its gradient, with torch.autograd.set_detect_anomaly(True).

Could you help me please?

IV_DQN works fine.

[Homry]

Hello. I have a similar error with the mujoco env (gym_walker 2d or gym_cheetah) and Duckietown env (gym_duckietown.simulator).

[manila95]

The code has been written particularly with the MBBL benchmark in mind which only supports torch==1.4.0. I'll try making a separate branch for Pybullet, mujoco env and Duckietown.

[AdityaGudimella]

Oh. I will try it with the MBBL benchmark and get back to you about it. Thank you!

[Homry]

Hello @hbutsuak95. Is there an approximate date when you will finish fixing the error?