Improving Autonomous Driving Safety with POP: A Framework for Accurate Partially Observed Trajectory Predictions

Hong Kong University of Science and Technology, RAM-LAB

Abstract & Method

Accurate trajectory prediction is crucial for safe and efficient autonomous driving, but handling partial observations presents significant challenges. To address this, we propose a novel trajectory prediction framework called Partial Observations Prediction (POP) for congested urban road scenarios. The framework consists of two key stages: self-supervised learning (SSL) and feature distillation. POP first employs SLL to help the model learn to reconstruct history representations, and then utilizes feature distillation as the fine-tuning task to transfer knowledge from the teacher model, which has been pre-trained with complete observations, to the student model, which has only few observations. POP achieves comparable results to top-performing methods in open-loop experiments and outperforms the baseline method in closed-loop simulations, including safety metrics. Qualitative results illustrate the superiority of POP in providing reasonable and safe trajectory predictions.

Case Analysis

We demonstrate a scenario in our simulator where a self-driving vehicle navigates through a congested traffic intersection, as shown in Figure \ref{fig}. During the initial phase of the simulation, the AV intends to traverse the intersection with a planned speed of 6.3 m/s. However, due to insufficient observation, the HiVT predictor inaccurately predicts the future trajectory in the first two frames. As a result, the AV fails to account for the movement of the vehicle below and begins to accelerate. By frame 3, the speed has already reached 8.7 m/s, making it too late to decelerate and leading to a collision. In contrast, the AV with the POP-H predictor consistently provides more reasonable predictions (indicated by the black scatter line) from frame 1 to frame 5, ensuring a higher level of safety.

More Animations

gif3

HiVT

gif4

POP-H

gif5

HiVT

gif6

POP-H

gif7

HiVT

gif8

POP-H

gif1

HiVT

gif2

POP-H

BibTeX

@misc{wang2023improving,
            title={Improving Autonomous Driving Safety with POP: A Framework for Accurate Partially Observed Trajectory Predictions}, 
            author={Sheng Wang and Yingbing Chen and Jie Cheng and Xiaodong Mei and Yongkang Song and Ming Liu},
            year={2023},
            eprint={2309.15685},
            archivePrefix={arXiv},
            primaryClass={cs.RO}
      }