Strangeness in Metrics
-
when calculating Jalone/Jcrowded, the angles are simply substracted (https://gitlab.inria.fr/CrowdBot/CrowdBotChallenge/-/blob/master/Crowdbotsim_python/crowdbotsimcontrol/metrics.py#L67). As a result, wrap-around (1 degree to 359 degree) leads to wrong values. A possible solution is to calculate angle differences instead of diff (https://gitlab.inria.fr/CrowdBot/CrowdBotChallenge/-/blob/flowbot/Crowdbotsim_python/crowdbotsimcontrol/metrics.py#L86)
-
when calculating the proximity metric (https://gitlab.inria.fr/CrowdBot/CrowdBotChallenge/-/blob/master/Crowdbotsim_python/crowdbotsimcontrol/metrics.py#L180), the loop for getting minimum relative distance resets the list every iteration. As a result, Proximity is a scalar (smallest distance at last timestep), but should be a list (smallest distance at every time step). (Also, instead of a loop, it can be done using amin instead https://gitlab.inria.fr/CrowdBot/CrowdBotChallenge/-/blob/flowbot/Crowdbotsim_python/crowdbotsimcontrol/metrics.py#L196)
Maybe there is another branch which contains fixed metric computation?