@andrey_zubkov: Did you get to figure out the issue? 52 steps means that the human model fell down in 0.52 s and total reward of 5 seems about right in such a case. The difference between your local environment and the server environment can be on the initial state where the muscle states can be slightly different when a simulation is initiated by
env.reset(...). So your controller should be robust enough to overcome the difference in init muscle state.
In addition, based on reports from other participants, it seems like there could have been more differences in the server evaluation. For Round 2, we only accept docker submission. So please try a docker submission and let us know if you still get drastically different results.
This issue is solved in the new version: https://github.com/stanfordnmbl/osim-rl/commit/dfb9ad48721bd93ed31a0eba876e29ed8c7d88ed
Sorry for the inconvenience. We will make sure the environments are consistent in Round 2.
This thread is to discuss the reward for Round 2, the final round. Hopefully, the discussion will help us to set a good reward by Oct 14, when Round 2 begins. Before sharing your concerns or asking for clarifications, please check out the following links:
FYI, self.LENGTH0 is there to normalize the velocities to model size (i.e. leg length) but is not relevant for the competition as it is set to self.LENGTH0 = 1.
Sorry for the delay. We are trying to send out the credits every weekend to those who newly appear in the ranking, but there seems to be some delay. We will send out those shortly. Please state in this thread when you receive them. Thanks.
Hi, @luisenp. The evaluation in Round 1 will be done with the same environment with the last version of the development environment. When you submit your solution, our server runs multiple simulations with different target velocity fields, and the evaluation score is the mean of the cumulative rewards you receive in those simulations.
@gupengju The problem should’ve been solved (let us know if it not). Also, note that now you have another way to submit your solution: https://www.aicrowd.com/organizers/stanford-neuromuscular-biomechanics-laboratory/challenges/neurips-2019-learn-to-move-walk-around#get-started
@huixxi You can find a new training example here: http://osim-rl.stanford.edu/docs/nips2019/training/
We will send out the $200 credit codes to those who qualified shortly
The evaluation environment will be with the dictionary with 4 keys.
@huixxi: Sorry for the confusion. You are correct and thanks for pointing it out. I corrected the document (http://osim-rl.stanford.edu/docs/nips2019/environment/#observation-input-to-your-controller)
Sorry about the confusion. We recently noticed this issue and am working on it so that the evaluation environment will give the same observation dictionary as the current local environment. We will let you know once this is solved.
You can change the physics parameters in the .osim file. For example, you can set zero gravity in the ./osim/models/gait14dof22musc_20170320.osim by changing line 9 to 0 0 0. Hope this helps.
We encourage participants with different background to collaborate and try interdisciplinary and novel approaches. Use this thread to call and search partners. For example, you can post something like this:
[Looking for deep RL experts]
Out current team: two gait biomechanics experts (“name 1” and “name 2”; homepage)
Goal: to win the “Most novel biomechanics solution” track
Contact: Email email@example.com to discuss more details