Robotics, Embodied AI, and Physical Computing research develops intelligent systems that are grounded in the physical world. We build robots, wearable technologies, and interactive devices that can perceive their surroundings, reason about action, learn from experience, and operate safely in complex real-world environments. Our work addresses a central challenge in modern AI: moving beyond systems that only process digital information, toward embodied agents that can understand physical scenes, manipulate objects, coordinate with people, and adapt as the world changes. By combining advances in artificial intelligence with sensing, computation, and physical interaction, this research supports novel technologies for healthcare, manufacturing, logistics, construction, agriculture, field robotics, rehabilitation, and assistive systems. It also contributes to the next generation of embodied foundation models: AI systems that integrate vision, language, action, touch, and physical dynamics to support more general, reliable, and adaptive behaviour.
Our research spans across multidisciplinary research
Robot learning is the engine that enables robots and physical AI systems to improve through experience – learning from data, demonstrations, simulation, and real-world interaction to perform tasks with greater speed, precision, and adaptability. By integrating robots with perception, control, and AI, robot learning helps machines understand their surroundings, respond to changing conditions, and master complex physical behaviours.
Our research investigates how robots can learn robust skills for manipulation, navigation, dexterous manipulation, long-horizon task execution, and human-guided operation. This includes imitation learning, reinforcement learning, self-supervised learning, learning from demonstration, and methods that integrate data-driven models with physical constraints, sensing, control, and human–machine interaction. A key focus is developing robot policies that generalise beyond controlled laboratory settings and remain reliable as objects, tasks, and environments vary.
Robot learning has significant potential to expand the range of tasks that robots can perform in real-world environments. By enabling robots to learn from human demonstration and adapt to new situations, this research supports applications in advanced manufacturing, logistics, healthcare, construction, agriculture, domestic assistance, and field robotics. It also reduces the engineering effort required to deploy robots, making robotic systems more flexible, scalable, and accessible across industries.
William Zhi, Professor Fabio Ramos, Dr Sasha Rubin, Associate Professor Tom Cai, Professor Zhiyong Wang
Wearable, haptic, and interactive physical systems integrate sensing, computation, feedback, and intelligent inference into devices that are worn on the body, embedded in objects, or placed within physical environments. These systems are a core part of physical computing, where intelligence is not confined to screens or servers but is embedded directly into the ways people move, work, sense, and interact.
Our research explores wearable sensing, haptic feedback, human–machine interfaces, assistive devices, and embodied AI systems that extend human capability through real-time sensing, feedback, and intelligent assistance. This includes technologies for understanding human movement, physiological state, activity, intention, and interaction with physical objects and environments. By combining physical devices with machine learning and real-time inference, we develop systems that can interpret human behaviour and provide timely, personalised, and context-aware support.
Wearable systems can improve how people interact with technology, workplaces, and physical environments. They support applications in rehabilitation, aged care, sports performance, workplace safety, human–robot collaboration, and assistive technologies for people with disabilities. By embedding intelligence into wearable and physical devices, this research enables more responsive, adaptive, and human-centred technologies.
Associate Professor Anusha Withana, Professor Eduardo Velloso
Planning and decision-making research develops methods that allow intelligent agents to select actions, reason about future outcomes, and coordinate behaviour over time. In robotics and embodied AI, planning is essential for connecting high-level goals with low-level physical actions, enabling systems to move safely, manipulate objects, complete multi-step tasks, and respond appropriately to changing environments.
Our research investigates task planning, motion planning, decision-making under uncertainty, multi-agent coordination, and integrated planning-and-learning systems. This includes methods that allow robots and embodied agents to reason over both symbolic task structure and continuous physical dynamics. We also study how systems can make decisions in uncertain, partially observed, and human-populated environments where safety, reliability, and adaptability are critical.
Planning and decision-making methods are central to deploying intelligent systems in real-world settings. They enable robots to perform complex tasks in homes, hospitals, warehouses, factories, construction sites, and public spaces. These methods also support autonomous systems that must act safely and efficiently under uncertainty, including mobile robots, collaborative robots, autonomous vehicles, and embodied AI assistants. By improving how machines reason about action and consequence, this research contributes to more trustworthy and capable intelligent systems.
William Zhi, Professor Fabio Ramos, Dr Sasha Rubin, Associate Professor Tom Cai, Professor Zhiyong Wang
Perception and scene understanding enable intelligent systems to interpret their surroundings from sensory data. This includes recognising objects, estimating geometry, understanding spatial relationships, tracking motion, interpreting human activity, and extracting task-relevant information from complex environments.
Our research develops methods for computer vision, 3D perception, multimodal sensing, scene representation, and semantic understanding. We investigate how systems can combine visual, depth, tactile, proprioceptive, and language-based information to build rich representations of the physical world. A central focus is enabling robots and embodied agents to move beyond passive recognition toward actionable understanding: knowing what objects are, where they are, how they can be used, and how they relate to current goals.
Perception and scene understanding are foundational for robots and intelligent systems that must operate outside controlled environments. These capabilities support applications in autonomous robotics, manufacturing, healthcare, infrastructure inspection, agriculture, logistics, augmented reality, and human–robot collaboration. By improving how machines understand physical scenes, this research enables safer, more capable, and more context-aware systems that can interact effectively with the real world.
William Zhi, Professor Fabio Ramos, Associate Professor Tom Cai, Professor Zhiyong Wang
World models are internal representations that allow intelligent systems to predict how the physical world will change in response to actions. In embodied AI, world models provide a foundation for reasoning, planning, simulation, and adaptation by helping agents understand not only what is present in a scene, but also what may happen next.
Our research develops world models that integrate vision, language, action, touch, spatial structure, object interaction, and physical dynamics. A rapidly emerging direction is the development of embodied foundation models: AI systems that combine multimodal perception with action-conditioned prediction to support more general and adaptive behaviour. These models can help robots anticipate consequences, detect when behaviour is likely to fail, and adapt before errors occur.
World models have the potential to make robots and embodied AI systems more reliable, data-efficient, and adaptable. By enabling agents to predict outcomes before acting, they can reduce trial-and-error learning, improve safety, and support more effective planning in complex environments. Applications include robotic manipulation, autonomous navigation, digital twins, industrial automation, assistive robotics, and simulation-based design. This research also contributes to broader advances in AI by grounding intelligence in physical prediction and interaction.
William Zhi, Professor Fabio Ramos, Associate Professor Tom Cai, Professor Zhiyong Wang
