Open lev-smolentsev-K1 opened 7 months ago
Hi @lev-smolentsev-K1
Sorry for the late reply. We were too busy to maintain the projects last month. And thanks for your interest.
The documentation bug is fixed already and merged into the main branch now. Thank you for pointing out this.
For map generation from config, yes, you can do that with the predefined PG blocks. Just take a look at how we create Multi-agent environments. The maps have fixed shape and structures. We are also working on the OSM/OpenDrive map, it is likely we can get it merged soon.
For trajectory planning, yes, MetaDrive can do that. As far as I know, some people are doing research with Casadi x MetaDrive and get some good results. But it might be hard to access future states as the future states of other objects will also be affected by current/future actions of ego car... But if the whole scenario is recorded like what we provide with scenarionet, it is feasible to get all future states of surrounding objects, which is indeed open-loop trajectory planning.
Hi guys, thanks for your work!
I've run into the same error in the last part of your tutorial (which I thought supposed to be working with the MyMapManager appearance) The error I'm talking is:
'NoneType' object has no attribute 'navi_arrow_dir'
(checked in both offline & collab)Secondly, a more general question: is it possible to generate roads from a sequence of parameters (like circles of various curvatures but not random pick) or might be one particular road like 'ABC' but for each word specify one value of the parameter (like length=100 for A, curvature = 5 for B and so on). Also a killing feature would be defining roads from OSM or OpenDRIVE descriptions (no need to generate anything as we have millions of km of road networks on the planet for example).
And lastly, I saw you provide bridges for ROS : is it possible to use your framework for a classic trajectory planning task e.g. when ego needs to track some planned trajectory updated at each step. Some classic planner should have access to the env state at current step but also at future steps (like trajectory prediction of obstacles or road signs states) is it possible to obtain these future states using MetaDrive (may be not during interactive simulation, e.g. recording a scenario and then playing it with ego and recorded obstacles in open loop interaction fashion)?
Thanks in advance for your answers!