Abstracts Track 2025

Area 1 - Intelligent Control Systems and Optimization

Nr:	41
Title:	Autonomous Multi-Agent Coordination for State-of-Charge Balancing in Electric Vehicles
Authors:	Yasser Bin Salamah
Abstract:	The rapid expansion of electric mobility, including cars, buses, and bikes, requires advanced battery management methods that are efficient, scalable, and resilient. Lithium-ion batteries (LiBs), recognized for their high energy density, cycle life, and reliability, remain the dominant choice for electric vehicles (EVs) [1], [2]. A major challenge, however, is maintaining uniform State of Charge (SoC) across cells, as variations due to internal resistance, degradation, and environmental factors can reduce driving range, shorten lifetime, and compromise safety. This work introduces an autonomous multi-agent coordination strategy for active SoC equalization in EV battery packs. Building on prior work in distributed control frameworks for wind energy and vehicular systems [3], [4], the proposed approach models each cell as an intelligent agent that communicates only with its neighbors. The framework integrates a multi-agent coordination algorithm with Proportional-Integral-Derivative (PID) compensation to achieve real-time balancing. A direct cell-to-cell topology with bidirectional switches and threshold-based logic enables efficient charge redistribution, eliminating the need for a central controller. Unlike conventional centralized balancing schemes, the method enhances scalability, improves fault tolerance, and ensures faster transient response. Simulation results demonstrate that initially imbalanced cells converge to a uniform SoC during both charging and discharging cycles. The coordination algorithm guarantees convergence toward the average SoC [5], while the PID action provides dynamic compensation and suppresses oscillations. By minimizing unnecessary switching and power loss, the method increases energy efficiency and reliability, making it highly suitable for EV applications where extended range, safety, and durability are essential. References: [1] N. Ghaeminezhad et al., “Active cell equalization topologies analysis for battery packs: A systematic review,” IEEE Trans. Power Electron., vol. 36, no. 8, pp. 9119–9135, 2021. [2] A. Kumar et al., “Battery management system in EV applications: Review, challenges and opportunities,” Materials Today: Proceedings, vol. 69, pp. 2451–2457, 2023. [3] Y. Bin Salamah and U. Ozguner, “Distributed extremum-seeking for wind farm power maximization using sliding mode control,” Energies, vol. 14, no. 4, p. 828, 2021. [4] A. Maarouf, Y. Bin Salamah, and I. Ahmad, “Decentralized control framework for optimal platoon spacing and energy efficiency,” Electronics, vol. 14, no. 1, p. 169, 2025. [5] R. Olfati-Saber, J. A. Fax, and R. M. Murray, “Consensus and cooperation in networked multi-agent systems,” Proc. IEEE, vol. 95, no. 1, pp. 215–233, 2007.

Area 2 - Robotics and Automation

Nr:	168
Title:	A Hybrid Semantic and Computer Vision Framework for Gesture Recognition and Intent Understanding on Pepper in Dementia Care
Authors:	Hiba Harrari
Abstract:	We present a novel hybrid architecture that integrates computer vision and semantic reasoning to improve gesture recognition and intent understanding in humanoid robots, specifically Pepper. While prior research has demonstrated the effectiveness of either semantic models or computer vision approaches, most systems adopt only one. However, this separation leaves a critical gap in contextual understanding: computer vision models may detect the gesture (e.g., pointing) but fail to interpret its intent (is it to indicate an object or to signal distress?), while semantic models lack the perceptual depth to handle visual complexity. This limitation is especially important in environments like dementia care, where gestures often carry emotional or urgent meanings. To the best of our knowledge, no existing system combines both approaches in a single unified framework for Pepper. Our system addresses this gap by combining 2D skeleton-based gesture recognition using the NTU RGB+D dataset and a state-of-the-art deep learning model with semantic reasoning layers that include a formal gesture ontology and Bayesian network inference. This hybrid architecture allows Pepper to not only recognize gestures with improved robustness and occlusion handling but also to infer the underlying intent and emotional significance behind those gestures. The proposed framework builds on skeleton extraction techniques inspired by Lefrant and Montanier’s work (“Online Skeleton Extraction and Gesture Recognition on Pepper”, ArXiv 2022), and semantic inference strategies adapted from Chiang et al.’s study on human activity recognition (“Culture as a Sensor?”, Int. J. of Social Robotics, 2019). Our system introduces a novel semantic-computer vision hybrid architecture designed to interpret these nuanced behaviors in context, allowing Pepper in the near future to recognize gestures and assist dementia patients accordingly by notifying caregiving staff, initiating calming interaction, or assisting in routine care, ultimately enhancing both patient safety and caregiver efficiency.