Reinforcement learning model agent trained from VR user studies data for Work Zone Safety III

This hdf5 file contains the weights for a reinforcement learning (RL) agent that calibrates/optimizes actions relating to a smartwatch alarm raised in response to hazardous vehicles near a roadway construction work zone:

1. Action 1: whether to raise a smartwatch alarms
2. Action 2: alarm modality (vibration/haptic pattern, or sound AND vibration/haptic pattern)
3. Action 3: alarm pattern characteristics (duration, frequency, and number of repetitions)

The RL agent is a Proximal Policy Optimization agent created using the Tensorforce library. More information can be found here: https://tensorforce.readthedocs.io/en/latest/agents/ppo.html

The weights are calculated from 500 training episodes based on smartwatch alarm reactions of participants in VR user testing of the Worker Safety III project experiments on an integrated traffic simulation and VR platform. All training episodes were based on data collected for Scenario 1: Placing cones in a mobile work zone only. More information on loading and saving weights can be found here: https://tensorforce.readthedocs.io/en/latest/basics/features.html#save-restore

For requesting the model and further questions, please contact Dr. Semiha Ergan at semiha@nyu.edu