-1

In the paper "Flow: Architecture and Benchmarking for Reinforcement Learning in Traffic Control", the authors said that "SUMO’scar following models are calibrated for a simulation timestep of 1.0 seconds, and their behavior for smaller timesteps is known to produce unnatural behaviors". However, I cannot understand why it will produce unnatural behaviors.

I have read the reference that the authors cited, "(2016) Simulation/basic definition. [Online]. Available: http://sumo.dlr. de/wiki/Simulation/Basic Definition#Defining the Time Step Length".

Yue Wang
  • 11

2 Answers2

1

I am not sure, guess the 1.0 second is related to human's reaction time. If the timestep is less than 1.0 seconds, it might be unrealistic.

Pao Raw
  • 37
  • 6
1

for a good answer to this question I recommend directly emailing the corresponding author of that paper! This is more of a general question than a software question and may require a substantive conversation. Hope that helps! It does have to do with what Pao Raw mentioned above.

Eugene Vinitsky
  • 56
  • 1
  • 3