Loading
Cookies help us deliver our services. By using our services, you agree to our use of cookies. Learn more

crowdAI is shutting down - please read our blog post for more information

NIPS 2017: Learning to Run

Reinforcement learning environments with musculoskeletal models


Completed
2154
Submissions
630
Participants
92467
Views

OsimEnv gets stuck on a particular step for a long time

Posted by << redacted >> over 2 years ago

Has anyone observed a case where the agent is in a particular position, and this position causes the OsimEnv to spend a long time (tens of seconds) computing that particular step(), before proceeding on? For me it was after a hop, where the foot looked like it was clipping the floor.

Has anyone found a workaround to speed up these particular “edge cases”?

2

Posted by NNAISENSE  over 2 years ago |  Quote

+1, nope

Posted by QinYongliang  over 2 years ago |  Quote

this is totally normal, for a constant accuracy solver. and yes, when forces are strong and nonlinear, longer time is needed per step.

very occasionally you will experience 5-10 minutes per step. There’s no way around it, except lower the accuracy, which can be done by modifying C source of opensim.

Posted by Prot  over 2 years ago |  Quote

I’m thinking of discouraging such cases by adding a penalty to the reward function :-)