DenseTNT: End-to-end Trajectory Prediction from Dense Goal Sets

By Junru Gu, Chen Sun, Hang Zhao. IEEE International Conference on Computer Vision, 2021

596 citations recorded by the source corpus.

Due to the stochasticity of human behaviors, predicting the future trajectories of road agents is challenging for autonomous driving. Recently, goal-based multi-trajectory prediction methods are proved to be effective, where they first score over-sampled goal candidates and then select a final set from them. However, these methods usually involve goal predictions based on sparse pre-defined anchors and heuristic goal selection algorithms. In this work, we propose an anchor-free and end-to-end trajectory prediction model, named DenseTNT, that directly outputs a set of trajectories from dense goal candidates. In addition, we introduce an offline optimization-based technique to provide multi-future pseudo-labels for our final online model. Experiments show that DenseTNT achieves state-of-the-art performance, ranking 1st on the Argoverse motion forecasting benchmark and being the 1st place winner of the 2021 Waymo Open Dataset Motion Prediction Challenge.

Subfields: Automotive Engineering, Safety, Risk, Reliability and Quality, Signal Processing

View Paper: https://www.semanticscholar.org/paper/70ed76ec9bff4eec17e6e0d023f85dffe760b9c1

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Automotive Engineering

In Safety, Risk, Reliability and Quality

In Signal Processing

Citations in the corpus

Query-Centric Trajectory Prediction (2023, methodology): “Compared with anchor-based methods that attempt to cover the ground truth with densely sampled handcrafted anchors [6, 19], our proposal module generates K adaptive anchors in a data-driven manner.”
A Survey on Trajectory-Prediction Methods for Autonomous Driving (2022, methodology): “…al. [139] propose a target-driven method called TNT, which deﬁnes sparse goal anchors and selects the best trajectory to the target, and DenseTNT [140] estimates dense goal candidates and get better results than TNT. Zeng et al. [141] use LaneRCNN to obtain the representation of each…”
Recent Advancements in End-to-End Autonomous Driving Using Deep Learning: A Survey (2023, methodology): “DenseTNT [43] It is a anchor-free model which directly outputs from the dense goal candidates GNN VectorNet VectorNet Goal Bases multi-trajectory prediction”
ST-P3: End-to-end Vision-based Autonomous Driving via Spatial-Temporal Feature Learning (2022, background): “Current motion prediction methods [36, 21, 27, 31] usually takes ground truth perception information and HD map as input, but they are susceptible to cumulative error when the perception input comes from other modules in real-life application.”
HiVT: Hierarchical Vector Transformer for Multi-Agent Motion Prediction (2022, methodology): “Using 82.1%, 74.6%, and 40.0% fewer parameters, HiVT-64 significantly outperforms LaneGCN [31], mmTransformer [32], and DenseTNT [22] on the metrics of minADE and minFDE.”
Motion Transformer with Global Intention Localization and Local Movement Refinement (2022, methodology): “VectorNet [16] is widely adopted by recent works [19, 43, 35, 47] due to its efficiency and scalability, where both road maps and agent trajectories are represented as polylines.”
DriveVLM: The Convergence of Autonomous Driving and Large Vision-Language Models (2024, background): “Specifically, 3D perception [28, 30, 39, 45] is limited to detecting and tracking familiar objects, omitting rare objects and their unique attributes; motion prediction [16, 18, 33, 35, 56] and planning [2, 31, 38] focus on trajectory-level actions, often neglecting the decision-level interactions…”