Weiheng Wang

Weiheng Wang

KIT

ActivePose: Active 6D Object Pose Estimation and Tracking for Robotic Manipulation

Published in arXiv (CoRR), 2025

Recommended citation: Sheng Liu, Zhe Li, Weiheng Wang, Han Sun, Heng Zhang, Hongpeng Chen, Yusen Qin, Arash Ajoudani, Yizhao Wang. (2025). "ActivePose: Active 6D Object Pose Estimation and Tracking for Robotic Manipulation." arXiv preprint arXiv:2509.11364. https://arxiv.org/pdf/2509.11364

ActivePose proposes an active 6-DoF object pose estimation and tracking system for robotic manipulation. It detects viewpoint-induced ambiguities online and resolves them by selecting next-best views (NBV) using a combination of VLM ambiguity prediction and FoundationPose entropy. For tracking under motion and occlusion, it introduces an equivariant diffusion-policy trained via imitation learning to generate camera trajectories that maintain object visibility and reduce ambiguity. :contentReference[oaicite:1]{index=1}

BibTeX

```bibtex @article{liu2025activepose, title = {ActivePose: Active 6D Object Pose Estimation and Tracking for Robotic Manipulation}, author = {Liu, Sheng and Li, Zhe and Wang, Weiheng and Sun, Han and Zhang, Heng and Chen, Hongpeng and Qin, Yusen and Ajoudani, Arash and Wang, Yizhao}, journal = {arXiv preprint arXiv:2509.11364}, year = {2025} }