google-deepmind / lab

A customisable 3D platform for agent-based AI research

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Problem found with agent's world position in game units (x,y,z)

jafphd opened this issue · comments

Hi, I found a problem with the player (agent)'s world position in game units (x,y,z). When I reset the environment using env.reset(), and then compute pos = env.observations()['DEBUG.POS.TRANS'], I get the defined specific "P" coordinate in the arena (see image). However, when running the simulation, for some reason, the agent coordinate starting position refers to the middle of the arena instead of "P" position.

I found a similar problem when the agent is, for example, teletransported to a predefined "P" starting position. However, by doing so, the agent's coordinates set to the middle of the arena every time (see further details).

I do not know why that is the case. It seems there is a mistake in the estimation on where the agent is placed or I am missing a setting on the starting position. Any comment on this issue would be appreciated. Thanks!

FURTHER DETAILS:

Starting point definition "P":

example2

Example path starting from the middle of the arena:

example3