Reproduce “DobotTrack: Target-Grounded Vision-Language-Action Learning for Embodied Person Tracking” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Shengjun Deng, Yinghan Zhang, Chengrui Wei, Liu Liang, Songyuan Liu, Pengkun Zheng, Yajun Zheng, Ying Zhang and 3 more

DobotTrack trains a robot policy to predict target masks and boxes alongside future waypoints, then uses target-conditioned queries to compress visual history. Under a fixed 16-token budget, target guidance raises Distracted Tracking success from 55.3% to 67.6%. Its 0.8B model averages 77.9% success across EVT-Bench’s three tasks, 16.9 points above reported OmTrackVLA results; this published-results comparison is not controlled.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.