Figures (10)  Tables (4)

    • Figure 1. 

      Recurrent neural network and long short-term memory cell structure.

    • Figure 2. 

      Structure of the Waymo motion dataset (uncompressed_scenario_validation_validation).

    • Figure 3. 

      Conceptual framework.

    • Figure 4. 

      Identification of lane changing behavior.

    • Figure 5. 

      Describing the driving buffer around the ego car. (a) Vehicle Operating Space in eight dimensions; (b) Vehicle Operating Space affected by surrounding vehicles.

    • Figure 6. 

      Training and testing loss during training process.

    • Figure 7. 

      Prediction accuracy and recall for training and testing datasets.

    • Figure 8. 

      Training loss decay for different feature combinations.

    • Figure 9. 

      Testing accuracy and recall for different prediction horizons.

    • Figure 10. 

      Testing accuracy and recall for different data balancing.

    • ReferenceMethodologyContextual information
      Deo et al.[20]LSTM, CNN, Social PoolingSurrounding Vehicles
      Hou et al.[22]LSTMSurrounding Vehicles
      Kim et al.[28]LSTMSurrounding Vehicles
      Liu et al.[52]Stacked TransformerHD map,
      Surrounding Vehicles
      Messaoud
      et al.[54,55]      		Attention, LSTM, Social PoolingSurrounding Vehicles
      Gao et al.[56]VectorNetHD map,
      Surrounding Vehicles
      Zhao et al.[57]LSTM, CNN, Social PoolingSurrounding Vehicles, Satellite Image
      Zhao et al.[57]
      Gu et al.[58]      		VectorNet, Goal-based PredictionHD map,
      Surrounding Vehicles
      Choi et al.[59]Attention, LSTM-
      Lin et al.[60]Attention, LSTM-

      Table 1. 

      Summary of deep learning in trajectory and behavior prediction.

    • HyperparameterValue
      Learning Rate0.005
      Number of Recurrent Layers1
      Number of features in hidden state64
      Batch Size32
      Number of Epochs100
      Threshold0.5
      Sequence Length4
      Selected FeaturesLongitudinal speed,
      Lateral Speed,
      Heading, VOS

      Table 2. 

      Hyperparameter setup for the vanilla model.

    • DatasetsNumber of observationAccuracyRecall
      Training Set18060.850.79
      Testing Set6020.810.75
      Validation Set6020.790.74

      Table 3. 

      Model performance on training, testing, and validation datasets.

    • Feature selectionFeature dimensionDatasetsAccuracyRecall
      Longitudinal Speed
      Lateral Speed
      Heading      		3Training Set0.810.80
      Testing Set0.790.79
      Validation Set0.770.75
      Longitudinal Speed
      Lateral Speed
      Heading
      Vehicle Coordinates      		5Training Set0.620.60
      Testing Set0.640.63
      Validation Set0.540.54
      Longitudinal Speed
      Lateral Speed
      Heading
      VOS      		11Training Set0.850.79
      Testing Set0.810.75
      Validation Set0.790.74

      Table 4. 

      Model performance for different feature combinations