search

vita-epfl / CrowdNav

[ICRA19] Crowd-aware Robot Navigation with Attention-based Deep Reinforcement Learning
MIT License
585 stars 169 forks source link

questions about 'agent.py' #8

Closed LeeKeyu closed 5 years ago

LeeKeyu commented 5 years ago

Hello! I am interested in your fantastic research work and have some questions when reading your code: In the 'agent.py' (/crowd_sim/envs/utils/), I noticed these lines in the 'get_next_observable_state' function: (when the kinematics is 'unicycle') next_vx = action.v np.cos(self.theta) next_vy = action.v np.sin(self.theta)

My question is why it should be 'self_theta' ? I mean why the next_vx and next_vy are not computed using the new theta = self.theta + action.r ? Could you please explain it? Thank you very much!

ChanganVR commented 5 years ago

Thank you for pointing out the mistake. After looking at the code, I think it's a bug and I have fixed it. However, this bug doesn't change any results in the paper because all current human policies only have the holonomic constraint.