Abstract:As a vital component in autonomous driving, accurate trajectory prediction effectively prevents traffic accidents and improves driving efficiency. To capture complex spatial-temporal dynamics and social interactions, recent studies developed models based on advanced deep-learning methods. On the other hand, recent studies have explored the use of deep generative models to further account for trajectory uncertainties. However, the current approaches demonstrating indeterminacy involve inefficient and time-consuming practices such as sampling from trained models. To fill this gap, we proposed a novel model named Graph Recurrent Attentive Neural Process (GRANP) for vehicle trajectory prediction while efficiently quantifying prediction uncertainty. In particular, GRANP contains an encoder with deterministic and latent paths, and a decoder for prediction. The encoder, including stacked Graph Attention Networks, LSTM and 1D convolutional layers, is employed to extract spatial-temporal relationships. The decoder is used to learn a latent distribution and thus quantify prediction uncertainty. To reveal the effectiveness of our model, we evaluate the performance of GRANP on the highD dataset. Extensive experiments show that GRANP achieves state-of-the-art results and can efficiently quantify uncertainties. Additionally, we undertake an intuitive case study that showcases the interpretability of the proposed approach. The code is available at
What problem does this paper attempt to address?
The problem that this paper attempts to solve is how to accurately predict the future trajectories of vehicles in autonomous driving while efficiently quantifying the uncertainty of the prediction. Specifically, although existing methods have made certain progress in capturing complex spatio - temporal dynamics and social interactions, there are problems of inefficiency and time - consuming when quantifying the prediction uncertainty. For example, many generative models quantify uncertainty by sampling from trained models, which is not only time - consuming but also inefficient. Therefore, this paper proposes a new model - the Graph Recurrent Attentive Neural Process (GRANP) model, aiming to solve these problems, achieve more accurate and efficient vehicle trajectory prediction, and be able to effectively quantify the prediction uncertainty.
### Model Overview
The GRANP model consists of an encoder and a decoder, where the encoder contains deterministic and latent paths, and the decoder is used for prediction. The specific structure is as follows:
- **Encoder**: It includes stacked Graph Attention Networks (GAT), Long Short - Term Memory Networks (LSTM), and 1D Convolutional Layers, which are used to extract spatio - temporal relationships.
- **Decoder**: It is used to learn the latent distribution, thereby quantifying the prediction uncertainty.
### Main Contributions
1. **Proposing the GRANP model**: The GRANP model combines the advantages of Neural Processes (NPs), can directly quantify and visualize prediction uncertainty, and shows robust and stable performance in various scenarios.
2. **Efficiently Capturing Spatio - temporal Relationships**: It uses GAT, LSTM, and triple 1D Convolutional Layers to efficiently capture spatio - temporal relationships.
3. **Performance Evaluation**: The performance of the model was evaluated on the highD dataset and compared with several state - of - the - art models. The results show that GRANP outperforms the baseline models in both prediction accuracy and uncertainty quantification.
### Experimental Results
- **Short - term Prediction**: In the prediction of the first 1 second, the performance of GRANP is slightly inferior to that of the PiP model.
- **Long - term Prediction**: In the prediction of the last 4 seconds, GRANP is significantly better than other baseline models. Specifically, GRANP reduces the prediction error by about 50% compared to the PiP model in terms of the RMSE metric, and has a 70% advantage in terms of the NLL metric.
- **Case Study**: By visualizing the prediction results in two common driving scenarios, lane - changing and going straight, it is shown that the predicted trajectories of the GRANP model are highly consistent with the real trajectories, and it can effectively quantify the uncertainty.
### Conclusion
The GRANP model performs excellently in the vehicle trajectory prediction task in autonomous driving. It not only improves the prediction accuracy but also can efficiently quantify the prediction uncertainty. These advantages make the GRANP model have important application value in complex driving scenarios.