EdgeVTP: Exploration of Latency-efficient Trajectory Prediction for Edge-based Embedded Vision Applications

TL;DR

EdgeVTP is a latency-efficient vehicle trajectory prediction method designed for edge devices, combining interaction-aware graph modeling with lightweight transformers to produce smooth predictions with predictable runtime.

Contribution

It introduces a novel approach that reduces decoding overhead and bounds interaction complexity, achieving state-of-the-art accuracy and lowest latency on highway benchmarks.

Findings

Achieves lowest end-to-end latency on multiple benchmarks.

Attains state-of-the-art prediction accuracy on two datasets.

Maintains competitive error rates on other benchmarks.

Abstract

Vehicle trajectory prediction is central to highway perception, but deployment on roadside edge devices necessitates bounded, deterministic end-to-end latency. We present EdgeVTP, an embedded-first trajectory predictor that combines interaction-aware graph modeling with a lightweight transformer backbone and a one-shot curve decoder. By predicting future motion as compact curve parameters (anchored at the last observed position) rather than horizon-scaled autoregressive waypoints, EdgeVTP reduces decoding overhead while producing smooth trajectories. To keep runtime predictable in crowded scenes, we explicitly bound interaction complexity via a locality graph with a hard neighbor cap. Across three highway benchmarks and two Jetson-class platforms, EdgeVTP achieves the lowest measured end-to-end latency under a protocol that includes graph construction and post-processing, while attaining state-of-the-art (SotA) prediction accuracy on two of the three datasets and competitive error on other benchmarks. Our code is available at https://github.com/SeungjinStevenKim/EdgeVTP.