search

openai / robosumo

Code for the paper "Continuous Adaptation via Meta-Learning in Nonstationary and Competitive Environments"
https://arxiv.org/abs/1710.03641
300 stars 80 forks source link

Possible Inconsistency between the Description and the Code? #3

Open gaoyuankidult opened 5 years ago

gaoyuankidult commented 5 years ago

https://github.com/openai/robosumo/blob/1862397b43990b663ea236cd77a56d0dd41c9f09/robosumo/envs/agents.py#L210

Hi,

When I read the paper, I found following text describing the observation of the agent to be

During the game, each agent observes positions of itself and the opponent, its own joint angles, the corresponding velocities, and the forces exerted on its own body (i.e., equivalent of tactile senses). (Paper) 4.2 Competitive (first paragraph)

However in the code, agent.py. get_obs function lines 190-212 indicates that the forces of the opponents are also included. Could it be that I read it wrong or some mistake?

Thanks a lot for your time and consideration on this matter!

Best,

@Danielhp95

alshedivat commented 5 years ago

Good catch. The code is correct -- the agent additionally observed forces exerted on the opponent's torso (not on any other limbs).

SigmaBM commented 5 years ago

Hi, @alshedivat I recorded the observations at every step, but found that observed forces exerted on agent itself and its opponent are always zero.