Perception
Human cues and multimodal learning
Visual representations of gaze, motion and interaction, with temporal modelling for intention and behavioural inference.
Machine learning · intelligent systems
Reliable intelligence in the real world.
I am Dr Samuel Adebayo. I study how intelligent systems perceive, reason and act when observations are incomplete and operating conditions change. My work spans computer vision, multimodal learning and embodied intelligence, alongside reliable LLM and agentic systems in production.
Research Fellow, Queen's University Belfast
Lecturer, Belfast Metropolitan College
Principal Data Scientist, ISx4
Research programme
One question, several settings.
Across physical and software systems, I am interested in the same underlying problem: how to make useful decisions from imperfect evidence without hiding uncertainty or failure modes.
Embodied systems
Perception that supports action
Learning, geometry and simulation for intelligent systems operating in physical environments, currently through robotic structural assembly.
Reliable AI
LLMs and agentic systems
Evaluation, observability and policy-constrained tool use, with human escalation where model autonomy should end.
Current research · ARISE
Robotic assembly as a testbed for reliable embodied intelligence.
I am a Research Fellow at Queen's University Belfast on ARISE, a US–Ireland programme involving Queen's, New York University, the University of Galway and the University of Texas at San Antonio. Within Queen's contribution, I lead machine-learning and computer-vision research for robotic structural assembly.
The work has produced ConPose and ISC-Perception, alongside experimental tooling that connects perception, simulation and physical interaction.
Selected work
Research outputs with a visible thread.
The papers below are deliberately selected to show the progression from human-centred perception to intelligent systems operating under real-world constraints.
Machine Vision and Applications · 2026
ConPose
Joint detection and pose learning for robotic assembly, designed around geometry-consistent supervision and deployment-facing evaluation.
Buildings · 2026
ISC-Perception
A hybrid real, photorealistic and synthetic vision dataset for near-field robotic assembly where real data are scarce.
News
Recent research activity.
Publications, project milestones and technical work. This is intentionally short; the full technical writing archive remains on samueladebayo.com.
September 2026
ISC-Perception selected for a Buildings cover
The visual highlights the hybrid data pipeline and robotic assembly setting behind the paper.
August 2026
ISC-Perception published in Buildings
Hybrid vision data for robotic assembly with novel Intermeshed Steel Connections.
June 2026
ConPose published in Machine Vision and Applications
A jointly trained single-pass framework for robotic perception.