Ziqi Lu

I am an Applied Scientist at the Frontier AI & Robotics team of Amazon. Before joining Amazon, I received my PhD from MIT CSAIL, under the supervision of Prof. John Leonard. During my PhD, I interned at Meta Reality Labs, Amazon AWS AI and NVIDIA Research.

My research focuses on robotic perception, 3D computer vision and machine learning. I am particularly interested in developing robust and object-aware robot perception and mapping systems.

I received my M.S.('19) degree from the University of California, Berkeley and B.S.('17) degree from Shanghai Jiao Tong University.

Email  /  GitHub  /  Google Scholar  /  LinkedIn

profile photo

News

[Feb 2025]   LoRA3D has been selected as a Spotlight in ICLR 2025 (scored in the top 2%) 🚀.

[Jan 2025]  Our work on LoRA3D: Low-Rank Self-Calibration of 3D Geometric Foundation Models has been accepted to ICLR 2025.

[Jan 2025]  Our work on 3DGS-CD: 3D Gaussian Splatting-based Change Detection for Physical Object Rearrangement has been accepted to IEEE RA-L.

[Dec 2024]  I joined the Frontier AI & Robotics team of Amazon as an Applied Scientist!

[Nov 2024]  I successfully defended my PhD thesis Addressing Challenges in Object-Based Robot Navigation and Mapping.

[Oct 2023]  I'll be joining the Autonomous Vehicle Research Group of NVIDIA as a research intern in Spring 2024!

[Apr 2023]  I'll be joining the Octant team of Amazon AWS AI as a research scientist intern this summer!

[Dec 2022]  Check out our Awesome-Object-SLAM repo for a list of object SLAM papers and resources.

[Jul 2022]   Our work on SLAM-Supported Self-Training for 6D Object Pose Estimation has been accepted to IROS 2022.

[Apr 2022]  Our work on Discrete-Continuous Smoothing and Mapping has been selected for the spotlight session (top 2 of 48) in the ICRA 2022 workshop on Robotic Perception and Mapping: Emerging Techniques.

[Jan 2022]  I'll be joining the Surreal team in the Meta Reality Labs as a research scientist intern this summer!

[Jul 2021]   Our work on Consensus-Informed Optimization Over Mixtures for Ambiguity-Aware Object SLAM has been accepted to IROS 2021.

Selected Publications

project image

LoRA3D: Low-Rank Self-Calibration of 3D Geometric Foundation Models


Ziqi Lu, Heng Yang, Danfei Xu, Boyi Li, Boris Ivanovic, Marco Pavone, Yue Wang
ICLR Spotlight , 2025
arxiv / website /

Self-specialization of 3D geometric foundation models (e.g. DUSt3R) to target scenes using sparse RGB images on a single GPU within just 5 min.

project image

3DGS-CD: 3D Gaussian Splatting-based Change Detection for Physical Object Rearrangement


Ziqi Lu, Jianbo Ye, John Leonard
RA-L, 2025
arxiv / code /

We estimate 3D object-level changes in 3DGS-represented scenes using sparse post-change images, within tens of seconds.

project image

Fast Sparse View Guided NeRF Update for Object Reconfigurations


Ziqi Lu, Jianbo Ye, Xiaohan Fei, Xiaolong Li, Jiawei Mo, Ashwin Swaminathan, Stefano Soatto
ArXiv, 2024
arxiv / website /

Update NeRFs under sparse view guidance to physical object reconfigurations.

project image

SLAM-Supported Self-Training for 6D Object Pose Estimation


Ziqi Lu, Yihao Zhang, Kevin Doherty, Odin Severinsen, Ethan Yang, John Leonard
IROS, 2022
arxiv / code /

A robust-pose-graph-optimization-aided self-training method for domain adaptation of 6D object pose estimators.

project image

Discrete-Continuous Smoothing and Mapping


Kevin Doherty, Ziqi Lu, Kurran Singh, John Leonard
RA-L, 2022
arxiv / video / code /

Selected for the spotlight session (top 2 of 48) in the ICRA 2022 workshop on Robotic Perception and Mapping: Emerging Techniques.

A novel solver to efficiently recover approximate solutions for discrete-continuous (hybrid) factor graph optimization problems in robotic applications.

project image

Consensus-Informed Optimization Over Mixtures for Ambiguity-Aware Object SLAM


Ziqi Lu*, QiangQiang Huang*, Kevin Doherty, John Leonard
IROS, 2021
paper / arxiv / video /

A real-time object-based SLAM system that is robust to symmetry- or occlusion-induced pose ambiguity from individual 6D object pose predictions.





Design and source code from Leonid Keselman's Jekyll fork of Jon Barron's website