industrialbenchmark icon indicating copy to clipboard operation
industrialbenchmark copied to clipboard

Published 20 hours ago • verified publisher icon siemens

Observations out of range

Open MHamza-Y opened this issue 2 years ago • 0 comments

https://github.com/siemens/industrialbenchmark/blob/ee195860c3b010d0d16be76ab2ca00d87b241a35/industrial_benchmark_python/IBGym.py#L100

Fatigue and consumption are upper bound to 1000. While they exceed value of 2000 in some cases. Which cause error when the observation is checked against the bounds by training algorithm in ray[rllib]

Dependencies tensorflow==2.9.1 industrial_benchmark_python==2.0 ray[all]==1.13.0 # rllib gym==0.21 torch==1.12.0+cu116 --extra-index-url https://download.pytorch.org/whl/cu116

Algorithm : PPO Industrial Benchmark: IBGym(70, reward_type='classic', action_type='continuous', reset_after_timesteps=1000)

Can be solved by increasing the upper bound.

MHamza-Y avatar Aug 01 '22 12:08 MHamza-Y