arXiv subjects

Marco Brocanelli

Publications and source records attributed to Marco Brocanelli.

7 recordsLinked to original sources

arXiv2026

WalkGPT: Grounded Vision-Language Conversation with Depth-Aware Segmentation for Pedestrian Navigation

Ensuring accessible pedestrian navigation requires reasoning about both semantic and spatial aspects of complex urban scenes, a challenge that existing Large Vision-Language Models (LVLMs) struggle to meet. Although these models can describe visual content, their lack of explicit grounding leads to object hallucinations and unreliable depth reasoning, limiting their usefulness for accessibility guidance. We introduce WalkGPT, a pixel-grounded LVLM for the new task of Grounded Navigation Guide, unifying language reasoning and segmentation within a single architecture for depth-aware accessibility guidance. Given a pedestrian-view image and a navigation query, WalkGPT generates a conversational response with segmentation masks that delineate accessible and harmful features, along with relative depth estimation. The model incorporates a Multi-Scale Query Projector (MSQP) that shapes the final image tokens by aggregating them along text tokens across spatial hierarchies, and a Calibrated Text Projector (CTP), guided by a proposed Region Alignment Loss, that maps language embeddings into segmentation-aware representations. These components enable fine-grained grounding and depth inference without user-provided cues or anchor points, allowing the model to generate complete and realistic navigation guidance. We also introduce PAVE, a large-scale benchmark of 41k pedestrian-view images paired with accessibility-aware questions and depth-grounded answers. Experiments show that WalkGPT achieves strong grounded reasoning and segmentation performance. The source code and dataset are available on the \href{https://sites.google.com/view/walkgpt-26/home}{project website}.

cs.CV↗

arXiv2025

Speeding-up Graph Algorithms via Clique Partitioning

Reducing the running time of graph algorithms is vital for tackling real-world problems such as shortest paths and matching in large-scale graphs, where path information plays a crucial role. To address this critical challenge, this paper introduces a graph restructuring algorithm that identifies bipartite cliques and replaces them with tripartite graphs. This restructuring leads to fewer edges while preserving complete graph path information, enabling the direct application of algorithms like matching and all-pairs shortest paths to achieve significant runtime reductions, especially for large, dense graphs. The running time of the proposed algorithm for a graph $G(V,E)$, with $|V| = n$ and $|E| = m$ is~$O(mn^\delta)$, which is better than $O(mn^\delta \log^2 n)$, the running time of the best existing algorithm for speeding-up other graph algorithms (the Feder-Motwani (\textsf{FM}) algorithm), where $0 \leq \delta \leq 1$. Both the \textsf{FM} algorithm and the proposed algorithm are originally formulated for bipartite graphs, but can also be applied to general directed or undirected graphs. Our extensive experimental analysis demonstrates that the proposed algorithm achieves up to 21.26\% higher reduction in the number of edges and runs up to 105.18$\times$ faster than the \textsf{FM} algorithm. On large synthetic graphs with up to 1.05 billion edges, it attains a reduction in the number of edges of up to 74.36\%. On real-world graphs, it achieves a reduction in the number of edges by up to 46.8\%. Furthermore, when used as a preprocessing step, our approach yields up to a 2.07$\times$ speedup for the matching algorithms on large synthetic graphs, and up to a 1.74$\times$ speedup for the All-Pairs Shortest Path algorithms on real-world graphs, when compared to using the given graph as input.

cs.DS↗

arXiv2025

Rethinking Energy Management for Autonomous Ground Robots on a Budget

Autonomous Ground Robots (AGRs) face significant challenges due to limited energy reserve, which restricts their overall performance and availability. Prior research has focused separately on energy-efficient approaches and fleet management strategies for task allocation to extend operational time. A fleet-level scheduler, however, assumes a specific energy consumption during task allocation, requiring the AGR to fully utilize the energy for maximum performance, which contrasts with energy-efficient practices. This paper addresses this gap by investigating the combined impact of computing frequency and locomotion speed on energy consumption and performance. We analyze these variables through experiments on our prototype AGR, laying the foundation for an integrated approach that optimizes cyber-physical resources within the constraints of a specified energy budget. To tackle this challenge, we introduce PECC (Predictable Energy Consumption Controller), a framework designed to optimize computing frequency and locomotion speed to maximize performance while ensuring the system operates within the specified energy budget. We conducted extensive experiments with PECC using a real AGR and in simulations, comparing it to an energy-efficient baseline. Our results show that the AGR travels up to 17\% faster than the baseline in real-world tests and up to 31\% faster in simulations, while consuming 95\% and 91\% of the given energy budget, respectively. These results prove that PECC can effectively enhance AGR performance in scenarios where prioritizing the energy budget outweighs the need for energy efficiency.

cs.RO↗

arXiv2023

GeoSAM: Fine-tuning SAM with Multi-Modal Prompts for Mobility Infrastructure Segmentation

In geographical image segmentation, performance is often constrained by the limited availability of training data and a lack of generalizability, particularly for segmenting mobility infrastructure such as roads, sidewalks, and crosswalks. Vision foundation models like the Segment Anything Model (SAM), pre-trained on millions of natural images, have demonstrated impressive zero-shot segmentation performance, providing a potential solution. However, SAM struggles with geographical images, such as aerial and satellite imagery, due to its training being confined to natural images and the narrow features and textures of these objects blending into their surroundings. To address these challenges, we propose Geographical SAM (GeoSAM), a SAM-based framework that fine-tunes SAM using automatically generated multi-modal prompts. Specifically, GeoSAM integrates point prompts from a pre-trained task-specific model as primary visual guidance, and text prompts generated by a large language model as secondary semantic guidance, enabling the model to better capture both spatial structure and contextual meaning. GeoSAM outperforms existing approaches for mobility infrastructure segmentation in both familiar and completely unseen regions by at least 5\% in mIoU, representing a significant leap in leveraging foundation models to segment mobility infrastructure, including both road and pedestrian infrastructure in geographical images. The source code can be found in this GitHub Repository: https://github.com/rafiibnsultan/GeoSAM.

cs.CV↗

arXiv2022

Are Turn-by-Turn Navigation Systems of Regular Vehicles Ready for Edge-Assisted Autonomous Vehicles?

Future private and public transportation will be dominated by Autonomous Vehicles (AV), which are potentially safer than regular vehicles. However, ensuring good performance for the autonomous features requires fast processing of heavy computational tasks. Providing each AV with powerful enough computing resources is certainly a practical solution but may result in increased AV cost and decreased driving range. An alternative solution being explored in research is to install low-power computing hardware on each AV and offload the heavy tasks to powerful nearby edge servers. In this case, the AV's reaction time depends on how quickly the navigation tasks are completed in the edge server. To reduce task completion latency, the edge servers must be equipped with enough network and computing resources to handle the vehicle demands. However, this demand shows large spatio-temporal variations. Thus, deploying the same amount of resources in different locations may lead to unnecessary resource over-provisioning. Taking these challenges into consideration, in this paper, we discuss the implications of deploying different amounts of resources in different city areas based on real traffic data to sustain peak versus average demand. Because deploying edge resources to handle the average demand leads to lower deployment costs and better system utilization, we then investigate how peak-hour demand affect the safe travel time of AVs and whether current turn-by-turn navigation apps would still provide the fastest travel route. The insights and findings of this paper will inspire new research that can considerably speed up the deployment of edge-assisted AVs in our society.

cs.NI↗

arXiv2021

Adversarially Robust and Explainable Model Compression with On-Device Personalization for Text Classification

On-device Deep Neural Networks (DNNs) have recently gained more attention due to the increasing computing power of the mobile devices and the number of applications in Computer Vision (CV), Natural Language Processing (NLP), and Internet of Things (IoTs). Unfortunately, the existing efficient convolutional neural network (CNN) architectures designed for CV tasks are not directly applicable to NLP tasks and the tiny Recurrent Neural Network (RNN) architectures have been designed primarily for IoT applications. In NLP applications, although model compression has seen initial success in on-device text classification, there are at least three major challenges yet to be addressed: adversarial robustness, explainability, and personalization. Here we attempt to tackle these challenges by designing a new training scheme for model compression and adversarial robustness, including the optimization of an explainable feature mapping objective, a knowledge distillation objective, and an adversarially robustness objective. The resulting compressed model is personalized using on-device private training data via fine-tuning. We perform extensive experiments to compare our approach with both compact RNN (e.g., FastGRNN) and compressed RNN (e.g., PRADO) architectures in both natural and adversarial NLP test settings.

cs.LG↗

arXiv2020

Long-Lived LoRa: Prolonging the Lifetime of a LoRa Network

Prolonging the network lifetime is a major consideration in many Internet of Things applications. In this paper, we study maximizing the network lifetime of an energy-harvesting LoRa network. Such a network is characterized by heterogeneous recharging capabilities across the nodes that is not taken into account in existing work. We propose a link-layer protocol to achieve a long-lived LoRa network which dynamically enables the nodes with depleting batteries to exploit the superfluous energy of the neighboring nodes with affluent batteries by letting a depleting node offload its packets to an affluent node. By exploiting the LoRa's capability of adjusting multiple transmission parameters, we enable low-cost offloading by depleting nodes instead of high-cost direct forwarding. Such offloading requires synchronization of wake-up times as well as transmission parameters between the two nodes which also need to be selected dynamically. The proposed protocol addresses these challenges and prolongs the lifetime of a LoRa network through three novel techniques. (1) We propose a lightweight medium access control protocol for peer-to-peer communication to enable packet offloading which circumvents the synchronization overhead between the two nodes. (2) We propose an intuitive heuristic method for effective parameter selections for different modes (conventional vs. offloading). (3) We analyze the energy overhead of offloading and, based on it, the protocol dynamically selects affluent and depleting nodes while ensuring that an affluent node is not overwhelmed by the depleting ones. Simulations in NS-3 as well as real experiments show that our protocol can increase the network lifetime up to $4$ times while maintaining the same throughput compared to traditional LoRa network.

cs.NI↗