Visualize single observation

[Syzygianinfern0]

How can I visualize the current_input: PlannerInput?

https://github.com/motional/nuplan-devkit/blob/cd3fd8d3d0c4d390fcb74d05fd56f92d9e0c366b/nuplan/planning/simulation/planner/simple_planner.py#L57

[Syzygianinfern0]

@christopher-motional could you please help me out if possible. Thanks!

[christopher-motional]

Hi @Syzygianinfern0, the devkit doesn't have anything out of the box for visualizing planner input directly, though perhaps you can visualize the scenarios the planner input is extracted from similar to what's being done in the scenario visualization tutorial. You can see here where the PlannerInput is constructed. The important part here is the history buffer which is an extraction of states from the scenario. Looking through the frames in the scenario should give you a picture of what you're interested in -- with a little work you could also try to visualize the history provided at each input by creating a similar method to serialize_scenario that creates a simulation history from past observations similar to what is being done when initializing a simulation history buffer from scenario.

[christopher-motional]

You can also see the visualization callback for visualizing features that get provided to MLPlanners. Hope that helps!

[Syzygianinfern0]

Thanks! Is there any existing method implemented to export simulations as mp4s?

[christopher-motional]

We don't support exporting simulations as mp4s, however you can use nuBoard (go to the Scenario tab) to visualize the scenarios associated with a set of simulations you generate much like a video file -- it also has a tab for quickly jumping between scenarios. You can see the latter part of the framework tutorial for an example of how to use it

[Syzygianinfern0]

perhaps you can visualize the scenarios the planner input is extracted from similar to what's being done in the scenario visualization tutorial. You can see here where the PlannerInput is constructed.

@christopher-motional this function requires a "scenario" object. How can I provide the input of the current_state and potentially the "next_state" so that I can visualize the movement of the ego vehicle for a single call of the planner.compute_trajectory()?

Also what is time_us?

[patk-motional]

It is easiest if you use the simulation framework.

Here is an example of running an open_loop simulation.

First, run the simulation. python nuplan/planning/script/run_simulation.py +simulation=open_loop_boxes planner=simple_planner

Inspect it using nuBoard. python nuplan/planning/script/run_nuboard.py