TrackVLA: Embodied Visual Tracking in the Wild

TL;DR

TrackVLA introduces a novel embodied visual tracking model that integrates recognition and trajectory planning using a shared language model, achieving state-of-the-art results in dynamic, occluded environments with high generalizability.

Contribution

The paper presents TrackVLA, a unified VLA model with a new benchmark and dataset for embodied visual tracking, enabling zero-shot generalization and robustness in real-world scenarios.

Findings

Achieves state-of-the-art performance on public benchmarks.

Demonstrates strong zero-shot generalization capabilities.

Operates at 10 FPS in complex environments.

Abstract

Embodied visual tracking is a fundamental skill in Embodied AI, enabling an agent to follow a specific target in dynamic environments using only egocentric vision. This task is inherently challenging as it requires both accurate target recognition and effective trajectory planning under conditions of severe occlusion and high scene dynamics. Existing approaches typically address this challenge through a modular separation of recognition and planning. In this work, we propose TrackVLA, a Vision-Language-Action (VLA) model that learns the synergy between object recognition and trajectory planning. Leveraging a shared LLM backbone, we employ a language modeling head for recognition and an anchor-based diffusion model for trajectory planning. To train TrackVLA, we construct an Embodied Visual Tracking Benchmark (EVT-Bench) and collect diverse difficulty levels of recognition samples, resulting in a dataset of 1.7 million samples. Through extensive experiments in both synthetic and real-world environments, TrackVLA demonstrates SOTA performance and strong generalizability. It significantly outperforms existing methods on public benchmarks in a zero-shot manner while remaining robust to high dynamics and occlusion in real-world scenarios at 10 FPS inference speed. Our project page is: https://pku-epic.github.io/TrackVLA-web.