Robust Inverse Rendering with Physics-Based Light Transport and Active Sensors
Abstract: Inverse rendering is the process of recovering the shape, materials, and lighting conditions of an environment from a set of images. Both this process as a whole and its individual components are fundamental to applications ranging from medical imaging to astronomy, and from AR/VR to embodied intelligence. In the thesis work discussed in this [...]
Communication Efficient and Differentially Private Optimization
Abstract: In modern machine learning, the abundance of data generated across diverse and distributed sources has made distributed training a central paradigm, particularly in large-scale applications such as Federated Learning. However, two key challenges arise in distributed training: ensuring communication efficiency and preserving the privacy of sensitive data used during training. This thesis addresses these [...]
Scaling in Robot Design for Locomotion and Manipulation
Abstract: The principles of scaling predict significant performance advantages for robots miniaturized to the millimeter and micrometer scales, including faster movements and higher precision. However, realizing these benefits is often hindered by manufacturing limitations and complex system-level interactions. This thesis investigates the effects of scaling on robot design for both manipulation and locomotion through the [...]
Scaling in Robot Design for Locomotion and Manipulation
Abstract: The principles of scaling predict significant performance advantages for robots miniaturized to the millimeter and micrometer scales, including faster movements and higher precision. However, realizing these benefits is often hindered by manufacturing limitations and complex system-level interactions. This thesis investigates the effects of scaling on robot design for both manipulation and locomotion through the development, fabrication, and characterization of [...]
Online-CBGT-Net: A Neuromimetic Architecture for Online Prediction
Abstract: We introduce Online-CBGT-Net, a neuromimetic architecture for interpretable, online prediction in streaming environments. Inspired by the cortico-basal ganglia-thalamic (CBGT) circuit in the mammalian brain, our model integrates evidence accumulation and a reset mechanism for continuous, online decision making. At each time step, the model receives input and incrementally accumulates evidence. A decision is made [...]
Toward Interactive Navigation in Unknown Dynamic Environments
Abstract: There is a growing demand for mobile robots to act not only as passive observers but also to actively interact with their environment, especially in cluttered and social settings. Meeting this demand requires navigation systems that can both understand interaction-relevant properties of the environment and make plans accordingly. This thesis presents a modular approach [...]
Learning Generalizable Robot Skills for Dynamic and Interactive Tasks
Abstract: Recent years have seen growing interest in developing robots capable of lifelong reliable operation in human-centric environments. Despite impressive recent progress towards long-horizon tasks such as laundry folding, current efforts are predominantly focused on quasi-static tasks in structured settings. General-purpose assistive robots should be capable of performing a wider range of dynamic and dexterous [...]
Resilient Autonomy for Extreme and Uncertain Environments
Abstract: Pushing the limit of robot autonomy in real-world challenging environments is the key to making it useful. One of the main challenges is to develop systems that are robust to extreme operating conditions, which introduces sensing degradation. In this talk, I'll present a series of works targeting extreme and uncertain environments, including scenes with dynamic [...]
Identifying Prompted Artist Names from Generated Images
Abstract: A common and controversial use of text-to-image models is to generate pictures by explicitly naming artists, such as “in the style of Greg Rutkowski”. Because the original prompt is usually unavailable, online platforms lack a reliable way to decide whether an uploaded image should be filtered for including artist names in its prompt---often required [...]
SplatSim: Zero-Shot Sim2Real Transfer of RGB Manipulation Policies Using Gaussian Splatting
Abstract: Sim2Real transfer, particularly for manipulation policies relying on RGB images, remains a critical challenge in robotics due to the significant domain shift between synthetic and real-world visual data. In this work, we propose SplatSim, a novel framework that leverages Gaussian Splatting as the primary rendering primitive to reduce the Sim2Real gap for RGB-based manipulation [...]