Zhiang Chen

Harish Anand, Stephen A. Rees, Zhiang Chen et al.

Simulations play a crucial role in robotics research and education. This paper presents the OpenUAV testbed, an open-source, easy-to-use, web-based, and reproducible software system that enables students and researchers to run robotic simulations on the cloud. We have built upon our previous work and have addressed some of the educational and research challenges associated with the prior work. The critical contributions of the paper to the robotics and automation community are threefold: First, OpenUAV saves students and researchers from tedious and complicated software setups by providing web-browser-based Linux desktop sessions with standard robotics software like Gazebo, ROS, and flight autonomy stack. Second, a method for saving an individual's research work with its dependencies for the work's future reproducibility. Third, the platform provides a mechanism to support photorealistic robotics simulations by combining Unity game engine-based camera rendering and Gazebo physics. The paper addresses a research need for photorealistic simulations and describes a methodology for creating a photorealistic aquatic simulation. We also present the various academic and research use-cases of this platform to improve robotics education and research, especially during times like the COVID-19 pandemic, when virtual collaboration is necessary.

Nahyuk Lee, Zhiang Chen, Marc Pollefeys et al.

Flow-matching methods for 3D shape assembly learn point-wise velocity fields that transport parts toward assembled configurations, yet they receive no explicit guidance about which cross-part interactions should drive the motion. We introduce TORA, a topology-first representation alignment framework that distills relational structure from a frozen pretrained 3D encoder into the flow-matching backbone during training. We first realize this via simple instantiation, token-wise cosine matching, which injects the learned geometric descriptors from the teacher representation. We then extend to employ a Centered Kernel Alignment (CKA) loss to match the similarity structure between student and teacher representations for enhanced topological alignment. Through systematic probing of diverse 3D encoders, we show that geometry- and contact-centric teacher properties, not semantic classification ability, govern alignment effectiveness, and that alignment is most beneficial at later transformer layers where spatial structure naturally emerges. TORA introduces zero inference overhead while yielding two consistent benefits: faster convergence (up to 6.9$\times$) and improved accuracy in-distribution, along with greater robustness under domain shift. Experiments on five benchmarks spanning geometric, semantic, and inter-object assembly demonstrate state-of-the-art performance, with particularly pronounced gains in zero-shot transfer to unseen real-world and synthetic datasets. Project page: https://nahyuklee.github.io/tora.

Zhiang Chen, Sarah Bearman, J Ramon Arrowsmith et al.

Robotic mapping is attractive in many scientific applications that involve environmental surveys. This paper presents a system for localization and mapping of sparsely distributed surface features such as precariously balanced rocks (PBRs), whose geometric fragility parameters provide valuable information on earthquake processes and landscape development. With this geomorphologic problem as the test domain, we carry out a lawn-mower search pattern from a high elevation using an Unpiloted Aerial Vehicle (UAV) equipped with a flight controller, GPS module, stereo camera, and onboard computer. Once a potential PBR target is detected by a deep neural network in real time, we track its bounding box in the image coordinates by applying a Kalman filter that fuses the deep learning detection with Kanade-Lucas-Tomasi (KLT) tracking. The target is localized in world coordinates using depth filtering where a set of 3D points are filtered by object bounding boxes from different camera perspectives. The 3D points also provide a strong prior on target shape, which is used for UAV path planning to closely map the target using RGBD SLAM. After target mapping, the UAS resumes the lawn-mower search pattern to locate and map the next target.