Search for: All records

Distribution networks, with large-scale integration of distributed renewable resources, particularly rooftop solar photovoltaic systems, represent the most extensive yet vulnerable components of modern electric power systems during climate extremes such as hurricanes. However, existing day-ahead electricity dispatch approaches primarily focus on the transmission network and lack the capability to manage the spatiotemporal risks associated with the vast distribution networks, which can potentially lead to significant power imbalances due to the mismatches between scheduled generation and actual demand. To address this increasingly critical gap under intensifying climate extremes and growing distributed renewable integration, we introduce Risk-aware Electricity Dispatch under Climate Extremes with Renewable integration (REDUCER), a risk-aware day-ahead electricity dispatch model that incorporates high-resolution spatiotemporal risk analysis for distribution networks with large-scale distributed renewable integration into an Entropic Value-at-Risk-constrained mixed-integer convex optimization framework. Applied to the 2022 Puerto Rico power grid under Hurricane Fiona, the proposed REDUCER model is seen to effectively manage these risks with substantially less reliance on additional flexibility resources to cope with power imbalances, reducing overall operational costs by about 30% under extreme cases compared to standard unit commitment strategies already informed by average demand loss. Also, the proposed REDUCER model consistently demonstrates its effectiveness in managing the increasing temporal net demand variability introduced by growing large-scale distributed solar integration while maintaining minimal operational costs. This model offers a practical solution for cost-effective and resilient electricity dispatch of modern power systems with large-scale renewable integration facing intensifying climate risks. 

Robot Imitation Learning (IL) is a crucial technique in robot learning, where agents learn by mimicking human demonstrations. However, IL encounters scalability challenges stemming from both non-user-friendly demonstration collection methods and the extensive time required to amass a sufficient number of demonstrations for effective training. In response, we introduce the Augmented Reality for Collection and generAtion of DEmonstrations (ARCADE) framework, designed to scale up demonstration collection for robot manipulation tasks. Our framework combines two key capabilities: 1) it leverages AR to make demonstration collection as simple as users performing daily tasks using their hands, and 2) it enables the automatic generation of additional synthetic demonstrations from a single human-derived demonstration, significantly reducing user effort and time. We assess ARCADE's performance on a real Fetch robot across three robotics tasks: 3-Waypoints-Reach, Push, and Pick-And-Place. Using our framework, we were able to rapidly train a policy using vanilla Behavioral Cloning (BC), a classic IL algorithm, which excelled across these three tasks. We also deploy ARCADE on a real household task, Pouring-Water, achieving an 80% success rate. 

Measuring and managing the risk of extensive distribution network outages during extreme events is critical for ensuring system-level energy balance in transmission network operations. However, existing risk measures used in stochastic optimization of power systems are computationally intractable for this problem involving large numbers of discrete random variables. Using a new coherent risk measure, Entropic Value-at-Risk (EVaR), that requires significantly less computational complexity, we propose an EVaR-constrained optimal power flow model that can quantify and manage the outage risk of extensive distribution feeders. The optimization problem with EVaR constraints on discrete random variables is equivalently reformulated as a conic programming model, which allows the problem to leverage the computational efficiency of conic solvers. The superiority of the proposed model is validated on the real-world Puerto Rico transmission grid combined with its large-scale distribution networks. 

While deep networks have achieved broad success in analyzing natural images, when applied to medical scans, they often fail in unexpected situations. This study investigates model sensitivity to domain shifts, such as data sampled from different hospitals or confounded by demographic variables like sex and race, focusing on chest X-rays and skin lesion images. The key finding is that existing visual backbones lack an appropriate prior for reliable generalization in these settings. Inspired by medical training, the authors propose incorporating explicit medical knowledge communicated in natural language into deep networks. They introduce Knowledge-enhanced Bottlenecks (KnoBo), a class of concept bottleneck models that integrate knowledge priors, enabling reasoning with clinically relevant factors found in medical textbooks or PubMed. KnoBo utilizes retrieval-augmented language models to design an appropriate concept space, paired with an automatic training procedure for recognizing these concepts. Evaluations across 20 datasets demonstrate that KnoBo outperforms fine-tuned models on confounded datasets by 32.4% on average. Additionally, PubMed is identified as a promising resource for enhancing model robustness to domain shifts, outperforming other resources in both information diversity and prediction performance. 

Abstract In this paper, we study a predator–prey mite model of Leslie type with generalized Holling IV functional response. The model is shown to have very rich bifurcation dynamics, including subcritical and supercritical Hopf bifurcations, degenerate Hopf bifurcation, focus‐type and cusp‐type degenerate Bogdanov–Takens bifurcations of codimension 3, originating from a nilpotent focus or cusp of codimension 3 that acts as the organizing center for the bifurcation set. Coexistence of multiple steady states, multiple limit cycles, and homoclinic cycles is also found. Interestingly, the coexistence of two limit cycles is guaranteed by investigating generalized Hopf bifurcation and degenerate homoclinic bifurcation, and we also find that two generalized Hopf bifurcation points are connected by a saddle‐node bifurcation curve of limit cycles, which indicates the existence of global regime for two limit cycles. Our work extends some results in the literature. 

Learning from Demonstration (LfD) is a powerful method for nonroboticists end-users to teach robots new tasks, enabling them to customize the robot behavior. However, modern LfD techniques do not explicitly synthesize safe robot behavior, which limits the deployability of these approaches in the real world. To enforce safety in LfD without relying on experts, we propose a new framework, ShiElding with Control barrier fUnctions in inverse REinforcement learning (SECURE), which learns a customized Control Barrier Function (CBF) from end-users that prevents robots from taking unsafe actions while imposing little interference with the task completion. We evaluate SECURE in three sets of experiments. First, we empirically validate SECURE learns a high-quality CBF from demonstrations and outperforms conventional LfD methods on simulated robotic and autonomous driving tasks with improvements on safety by up to 100%. Second, we demonstrate that roboticists can leverage SECURE to outperform conventional LfD approaches on a real-world knife-cutting, meal-preparation task by 12.5% in task completion while driving the number of safety violations to zero. Finally, we demonstrate in a user study that non-roboticists can use SECURE to efectively teach the robot safe policies that avoid collisions with the person and prevent cofee from spilling.