Are reactive agents working reliably in closed-loop simulations?

[d451gon]

I've just started running some simulations with the SimplePlanner to test the different settings, and I'm getting very odd behaviors when using reactive agents in closed-loop simulations (see gifs below):

Using the latest release (1.1). Code to generate the first gif:

python run_simulations.py +simulation=closed_loop_reactive_agents scenario_filter.limit_total_scenarios=10 scenario_filter.timestamp_threshold_s=70

The first gif corresponds to the only scenario of type "near_multiple_vehicles" that results from running the code above (note that this is from the log 2021.07.16.20.45.29_veh-35_01095_01486, in the mini dataset, as per the default config).

The second gif is from a different log, just to double check.

Additionally, the number of vehicles surrounding the ego-vehicle in these simulations appears to be suspiciously low, although this might be a coincidence.

Thanks in advance for any insights!

[patk-motional]

Hi @d451gon,

I tried it myself but I don't seem to be able to recreate the issue. I was able to generate the gif below.

[d451gon]

Same as in #254 downgrading numpy from 1.24.2 to 1.23.4 solved the issue.

Thanks @patk-motional !