Designing for Physical AI

Physical AI refers to AI systems that sense and act in the physical world rather than only on a screen: robots, autonomous machines, wearables, voice assistants, and ambient or embodied devices. This course approaches that category as a design problem rather than an engineering one. Over five weeks you work through the three dimensions that shape an embodied experience, the people using it, the place it operates in, and the technology carrying it, and you apply each to a real product through three design challenges. You leave able to select sensory modes for a context, evaluate privacy and ethics tradeoffs, assess a physical environment, and communicate an interaction concept for technology that has no design precedent.

The label is contested, and fairly so. It was popularised by chip vendors and has been stretched to cover everything from robot vacuums to a language model with tool access, which is a reasonable thing to be sceptical about. What this course teaches does not depend on the term surviving. Human factors, environmental psychology, inclusive design, and multimodal interaction are established disciplines, and the classes are grounded in design practices and frameworks that can be applied to roles that exist today, with one dedicated week extending into speculative, future-facing design. If the category ends up being called embodied systems or ambient computing instead, the frameworks move with you.

No. There is no coding in this course and no engineering prerequisite. What you need is a basic understanding of, or interest in, user experience principles, and comfort engaging in weekly studio critique and in class collaborative exercises. The course teaches you to make and defend design decisions about Physical AI systems, which is a different skill from building them. The three design challenges are analytical and conceptual: a journey map, an environmental assessment, and a product concept.

The loop you rely on disappears. You cannot ship, measure, and patch your way to a good answer when the thing is a device in someone’s kitchen or a robot crossing a floor, and you cannot put a tooltip on a haptic buzz. Week 1 has you tear down a shipping product to see those choices already made. Week 3 covers environmental psychology, because the room is now part of the interface. Week 4 is the mode decision itself: whether an interaction should be visual, auditory, haptic, or ambient, and which device it belongs on. Those are the three things screen practice does not prepare you for.

Almost every Physical AI course available teaches how these systems work: simulation, robotics stacks, model training, digital twins. This one teaches how to design them. You spend the five weeks on sensory mode selection, environmental context, ethics and privacy tradeoffs, and interaction concepts for technologies with no design precedent yet. If you are trying to train a policy in Isaac Sim, the technical courses are the right choice. If you are trying to decide whether a system should speak, buzz, glow, or stay silent in someone’s kitchen, that question is this course.

By anchoring the speculative work to something real, which is why the challenges run in this order. You start from a product that already ships and take it apart. You assess an actual space and the behaviour it produces. Only then do you design forward, and the Week 5 concept is grounded against the case study you have been building since Week 1. Journey mapping and scenario based storytelling are the tools that make a concept legible to stakeholders who cannot try it yet. Two of the three design challenges are analysis of real products and real places, not imagination.

That judgement is Week 4. The course treats mode selection as a decision you have to argue for: which of visual, auditory, haptic, or ambient fits a given physical and social context, and what the tradeoffs are between explicit interfaces like voice and chat and implicit, ambient interactions. Voice is often the wrong answer, and being able to say why, in terms of the room, the people in it, and what the system would have to interrupt, is more useful than knowing how to design a good prompt. You leave with a framework for making that call rather than a default.

Bystander consent is one of the hardest open problems in this space, and the course treats it as a design question rather than a legal footnote. Week 2 is an applied privacy and ethics audit of a real product, alongside inclusive design and the tradeoffs that come with it. Week 3 looks at the relationship between place, people, and technology, including public and private spaces. The person a camera captures in a lobby is not someone you can recruit, consent, or debrief, and the course gives you a structured way to surface that in a design review before the product ships rather than after.

You leave with one end-to-end case study built from three connected challenges: a journey map of a current Physical AI product experience, an environmental assessment of a real space with technology already in it, and a future-facing concept grounded against the product you tore down in Week 1. The concern about speculative work is legitimate, and the structure answers it. Only the final challenge is future-facing, and it is anchored to two pieces of analysis of things that actually exist. What a reviewer sees is your reasoning about a real product and a real place, then where you took it, which is a different artefact from a redesign invented for a portfolio.

All three, and each seat gets something different. Designers get hands-on practice designing an embodied AI system and a framework for choosing sensory modes that build trust instead of eroding it. Researchers build methods for a genuinely new research problem: evaluating a system that is simultaneously building a model of the user while the user is building a model of it. Product managers and strategists practise the discussion that sits above any single design decision, including where a product should land on the current to speculative spectrum and what a system is allowed to attempt and fail at before it damages trust. The three design challenges give every role a shared case study to work through together.

Two hours live each Monday for five weeks, plus time between sessions for the design challenges. Each challenge is kicked off in class, so you are never starting from a blank page alone. Every session is recorded and posted, and lifetime access to recordings and materials is included, though participation and student-led discussion are weekly, so the live sessions are where most of the value sits.

Yes. Certificate completion is based on submitting the three design challenges and attending and participating in class discussion. Participation and student-led discussion are weekly, so the certificate reflects work you actually did rather than attendance alone.