Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

TPK: Trustworthy Trajectory Prediction Integrating Prior Knowledge For Interpretability and Kinematic Feasibility

Marius Baden, Ahmed Abouelazm, Christian Hubschneider, Yin Wu, Daniel Slieter, J. Marius Zöllner

TL;DR

This work proposes incorporating interaction and kinematic priors of all agent classes-vehicles, pedestrians, and cyclists with class-specific interaction layers to capture agent behavioral differences and introduces DG-SFM, a rule-based interaction importance score that guides the interaction layer.

Abstract

Trajectory prediction is crucial for autonomous driving, enabling vehicles to navigate safely by anticipating the movements of surrounding road users. However, current deep learning models often lack trustworthiness as their predictions can be physically infeasible and illogical to humans. To make predictions more trustworthy, recent research has incorporated prior knowledge, like the social force model for modeling interactions and kinematic models for physical realism. However, these approaches focus on priors that suit either vehicles or pedestrians and do not generalize to traffic with mixed agent classes. We propose incorporating interaction and kinematic priors of all agent classes--vehicles, pedestrians, and cyclists with class-specific interaction layers to capture agent behavioral differences. To improve the interpretability of the agent interactions, we introduce DG-SFM, a rule-based interaction importance score that guides the interaction layer. To ensure physically feasible predictions, we proposed suitable kinematic models for all agent classes with a novel pedestrian kinematic model. We benchmark our approach on the Argoverse 2 dataset, using the state-of-the-art transformer HPTR as our baseline. Experiments demonstrate that our method improves interaction interpretability, revealing a correlation between incorrect predictions and divergence from our interaction prior. Even though incorporating the kinematic models causes a slight decrease in accuracy, they eliminate infeasible trajectories found in the dataset and the baseline model. Thus, our approach fosters trust in trajectory prediction as its interaction reasoning is interpretable, and its predictions adhere to physics.

TPK: Trustworthy Trajectory Prediction Integrating Prior Knowledge For Interpretability and Kinematic Feasibility

TL;DR

This work proposes incorporating interaction and kinematic priors of all agent classes-vehicles, pedestrians, and cyclists with class-specific interaction layers to capture agent behavioral differences and introduces DG-SFM, a rule-based interaction importance score that guides the interaction layer.

Abstract

Trajectory prediction is crucial for autonomous driving, enabling vehicles to navigate safely by anticipating the movements of surrounding road users. However, current deep learning models often lack trustworthiness as their predictions can be physically infeasible and illogical to humans. To make predictions more trustworthy, recent research has incorporated prior knowledge, like the social force model for modeling interactions and kinematic models for physical realism. However, these approaches focus on priors that suit either vehicles or pedestrians and do not generalize to traffic with mixed agent classes. We propose incorporating interaction and kinematic priors of all agent classes--vehicles, pedestrians, and cyclists with class-specific interaction layers to capture agent behavioral differences. To improve the interpretability of the agent interactions, we introduce DG-SFM, a rule-based interaction importance score that guides the interaction layer. To ensure physically feasible predictions, we proposed suitable kinematic models for all agent classes with a novel pedestrian kinematic model. We benchmark our approach on the Argoverse 2 dataset, using the state-of-the-art transformer HPTR as our baseline. Experiments demonstrate that our method improves interaction interpretability, revealing a correlation between incorrect predictions and divergence from our interaction prior. Even though incorporating the kinematic models causes a slight decrease in accuracy, they eliminate infeasible trajectories found in the dataset and the baseline model. Thus, our approach fosters trust in trajectory prediction as its interaction reasoning is interpretable, and its predictions adhere to physics.
Paper Structure (27 sections, 6 equations, 3 figures, 5 tables)

This paper contains 27 sections, 6 equations, 3 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Interaction Priors
  4. Kinematic Priors
  5. Methodology
  6. Interpretable Encoder
  7. Class-Specific Agent-to-Agent Layer
  8. Interaction Prior
  9. Interaction Prior Integration
  10. Multiply-and-Renormalize (MnR)
  11. Gating-and-Loss (GnL)
  12. Feasible Decoder
  13. Kinematic Priors
  14. Kinematic Prior Integration
  15. Experiment Setup
  16. ...and 12 more sections

Figures (3)

  • Figure 1: Comparison of learned agent attention in HPTR and our proposed model on a scene from the ArgoVerse 2 dataset. The focal agent is highlighted in cyan, with numbers indicating the attention score per agent and black arrows representing agent velocities.
  • Figure 2: The scene is first encoded in agent and map element embeddings and enhanced by transformer layers. Then, the prior-integrated and class-specific agent-to-agent layer captures interactions between agents. Next, transformer layers generate mode-specific embeddings based on the agent embeddings. Subsequently, class-specific heads predict a sequence of control inputs and a confidence value per mode. Finally, a kinematic layer deterministically calculates the predicted trajectories resulting from the control inputs.
  • Figure 3: The reachable area within a single time step when constraining a pedestrian walking with kinematic models. Among these, we find that the double integrator best represents real-world behavior.