
Jelin Raphael Akkara
PhD Researcher in Embodied AI
University of Padua / Fondazione Bruno Kessler
About
I am pursuing a PhD in Brain, Mind, and Computer Science at the University of Padua, in collaboration with Fondazione Bruno Kessler. My research in Embodied AI explores how autonomous agents ground human goals in complex environments and adapt when actions fail. My interests include fine-grained object navigation, personalized spatial reasoning, and memory-guided adaptation, with a growing focus on failure understanding and recovery.
My current work investigates how agents can use multimodal reasoning and memory to diagnose failures, reason about alternative actions, and retrieve past experiences to guide recovery. I am also interested in how causal reasoning and world models can help agents anticipate failures and evaluate the consequences of their actions.
Publications
IJCV 2026
Personal-3D: A Comprehensive Benchmark for Personalized Embodied AI Agents
Filippo Ziliotto, Jelin Raphael Akkara, Alessandro Daniele, Lamberto Ballan, Luciano Serafini, Tommaso Campari
Experience
Teaching Assistant - Computer Vision and Cognitive Systems
Assisted with lab sessions, guided students on assignments and projects, and facilitated discussions on advanced topics in computer vision and cognitive systems.
GenAI Research Intern
ViWear
Enhanced a Stable Diffusion-derived virtual try-on model, improving perceptual metrics by 7-12% and tripling inference speed from 1 fps to 3 fps through architectural improvements and GPU-constrained fine-tuning.
Computer Vision Research Intern
Fondazione Bruno Kessler / Visual Intelligence and Machine Perception Group, University of Padua
Guided by: Prof. Lamberto Ballan and Tommaso Campari
Enhanced scene mapping with multimodal vision-language features while keeping the pipeline memory-efficient, increasing navigation success by 7.6% and reducing memory usage by 99.75%.
Education
University of Padua / Fondazione Bruno Kessler
PhD in Brain, Mind, and Computer Science (BMCS)
Research focus: Building embodied agents that use multimodal reasoning and memory to understand their environments, adapt to users, and learn to anticipate and recover from failures.
University of Padua, Italy
MSc in Physics of Data
Thesis: Zero-Shot Object Goal Navigation using Online Image Retrieval
GPA: 110 / 110
National Institute of Technology, Calicut, India
B.Tech in Engineering Physics
First Class with Distinction

