From computer vision to autonomous robotics, artificial intelligence is moving into the physical world, changing how organizations observe their environments, make decisions and act.
Generative AI changed expectations by showing how machines can create text, images and code. Physical AI pushes the idea further: systems can perceive their surroundings, reason about what they observe, decide what information they need next and take action in the real world.
This convergence of artificial intelligence, computer vision, robotics, sensors and simulation is often called Physical AI. The important distinction is not the hardware itself. It is the shift from systems that only analyze information they are given to systems that can actively gather information, interpret it and adapt their behaviour to achieve an objective in complex industrial, logistics, medical or urban environments.
For businesses, this shift opens a different class of opportunities.. Physical AI solutions can automate complex inspections, strengthen quality-control processes, enhance infrastructure monitoring and reduce workers’ exposure to repetitive or hazardous tasks
But technical capability alone does not create business value. The strongest Physical AI initiatives start with an operational problem, a measurable outcome and a clear understanding of how people will work with the system. The technology should serve the strategy, not the other way around.
“Physical AI gives artificial intelligence eyes to observe, the ability to reason and decide, and the means to act in the real world.” - Martin Coulombe, CEO of Osedea
Physical AI: Grounded in the Real World
Most AI systems deployed to date rely on information that has already been collacted, such as documents, databases, images or transaction histories.
Physical AI changes the relationship with data. Instead of only receiving information, systems can actively participate in
Consider a fixed camera. It records what is in front of it from a predetermined angle. A Physical AI system can go further: it can determine another angle is needed, reposition a sensor or mobile platform, compare multiple perspectives and gather additional evidence before reaching a conclusion. . The system could therefore decide to gather additional information before reaching a conclusion based on a more complete observation. The resulting data becomes more targeted, contextual and directly connected to the operational objective.
That shift, from passive observation to purposeful perception, is what makes more autonomous and adaptable systems possible.
-
In a factory, a mobile robot can independently inspect multiple areas within minutes and adjust its route based on the equipment and issues it encounters.
-
In public infrastructure, a robot could document site conditions and report anomalies.
-
In a healthcare environment, a connected device could support data collection while simplifying the work of professionals.
For businesses, adopting physical AI can deliver several tangible benefits:
-
More accurate resource planning through real-time, contextual data about equipment, operations, and on-site needs.
-
Greater operational efficiency by automating repetitive, complex, or time-consuming tasks, allowing teams to focus on higher-value activities.
-
Improved foresight through early anomaly detection, predictive maintenance, and fewer unplanned disruptions.
-
Faster, better-informed decisions based on targeted, contextualized data directly connected to operational realities.
From Repetitive Execution to Intelligent Autonomy
Traditional industrial automation relies on repeatability: a robotic arm follows a predefined path and reproduces the same sequence with precision. This approach remains extremely useful, but it quickly reaches its limits when components, surfaces or conditions change. Each new situation then requires additional programming.
Physical AI introduces a different logic. Instead of programming every movement, organizations can provide the system with a clear objective and the constraints associated with achieving it. The system whether a robot, sensor, camera or other physical device can then adapt its data collection, select a new perspective or modify its sequence of actions based on what it observes, maximizing its ability to achieve the predetermined objective.
Greater autonomy, however, requires more engineering discipline, not less. Osedea combines agentic reasoning with deterministic safety protocols. Simulation and digital twins also make it possible to test behaviours, reproduce edge cases and validate decisions in a risk-free environment before deployment in the field. Moving from simulation to the physical world is therefore not a final handoff; it is a critical engineering stage in any Physical AI project.
What Physical AI looks like in practice
Physical AI should not be judged by how futuristic a system looks. Its value is measured by the problem it solves and how well it performs in the environment where people actually need it. Across manufacturing, public infrastructure and healthcare, Osedea's work shows what happens when AI, computer vision, software and robotics are engineered around real operational constraints.
Smarter manufacturing inspections
For Promark Electronics, Osedea developed KonnectAi, a quality-control system that uses computer vision to facilitate visual inspections during production and at the end of the manufacturing process. A camera captures images of components, and AI models analyze specific criteria, such as the presence of screws or the conformity of connections.
The system retains timestamped images to ensure traceability and allows inspectors to confirm a defect or report a false positive. This feedback supports the continuous improvement of the models. The project demonstrates that an effective solution depends on more than detection alone: it must also integrate into existing workflows, support human intervention and evolve as components and operational needs change.
An Autonomous Robot Inspection in the Montréal Metro
As part of an experimental project with the Société de transport de Montréal, Osedea used the Spot robot to conduct autonomous inspections at Bonaventure station outside operating hours. The robot captured images to independently identify certain anomalies, including waste, stickers, graffiti and defective light bulbs.
Osedea also designed the application used to organize and analyze the collected data. The project explored more than robotic mobility: it examined how a mobile system can navigate a complex environment, turn large volumes of images into structured information and help staff identify necessary interventions. Over time, that historical data can also reveal patterns that support better operational planning.
Extending existing tools in healthcare
Working with Halo Dental Technologies, Osedea contributed to the development of a connected dental mirror and medical-records platform combining computer vision, natural language processing, software and hardware. The integrated camera transmits images to a tablet, while AI can support functions such as image enhancement and voice transcription of clinical observations.
Physical AI does not have to look like a robot. It can be embedded in a familiar professional tool and make that tool more capable. In healthcare, that means designing intelligence around the people using it, while incorporating security, privacy and regulatory requirements from the outset. The goal is not to replace professional judgment, but to reduce friction and extend what the tool can help a professional accomplish.
Moving from a Pilot Project to Measurable Value
A compelling demonstration proves that a technology can work. It does not prove that it will create value in day-to-day operations. A Physical AI initiative becomes meaningful when it solves a clearly defined operational problem, works within the realities of the environment and can be governed, maintained and improved over time. Four questions help separate a promising experiment from a deployable system:
-
What business outcome are we seeking to achieve? Whether the objective is to reduce inspection time, improve quality, strengthen traceability, enhance safety or increase equipment availability, success must be measured against a tangible result.
-
How will the solution integrate into existing operations? Interfaces, data, human responsibilities, maintenance requirements and escalation processes must be designed alongside the model or robot.
-
How will autonomy be governed? Permissions, safety protocols, simulation-based validation and human intervention must be proportional to the level of risk.
-
How could the solution evolve? A sustainable architecture should allow new models, sensors and use cases to be added without multiplying costs and complexity.
There is another important lesson: autonomy does not remove people from the equation. In many cases, it makes their expertise more important during design and validation. Operators, inspectors and professionals understand the edge cases, constraints and context that technology alone cannot reproduce. Physical AI is most effective when that field expertise shapes the need, the system's behaviour and how the solution improves over time.
Canada Has the Ingredients. Deployment Is the Test.
Canada has many of the ingredients needed to lead in Physical AI: recognized AI expertise, a diverse industrial base, world-class researchers and companies capable of bringing software, data, design and engineering together. The opportunity now is to convert that capability into systems that work reliably in factories, infrastructure, logistics networks and healthcare environments. Research strength matters; deployment is where economic and operational value is created.
For organizations, the objective should not be maximum automation. It should be better operations. The highest-value opportunities are often the places where perception, targeted data collection and autonomous action can remove a bottleneck, improve a decision, strengthen safety or free people to focus on higher-value work.
The organizations that lead will not be the ones that apply AI to the greatest number of processes. They will be the ones that identify where intelligence can materially change an outcome, experiment with purpose and engineer for integration and scale from the beginning. Physical AI is already moving AI beyond the screen.
Meet Osedea at ALL IN 2026
On September 16 and 17, 2026, ALL IN will bring together more than 7,500 leaders from 40 countries in Montréal. Physical AI and robotics will feature prominently in discussions about the technologies transforming our industries and economy.
Osedea will participate in ALL IN 2026 with a dedicated booth and a presentation on the ALL IN Talks stage. Meet their team to discover practical, real-world AI applications, discuss your operational challenges and explore how to turn an ambitious idea into a tangible solution that will help you achieve your business objectives.
About Osedea
Osedea is a Montréal-based, engineering-led innovation firm that helps organizations turn ambitious ideas into intelligent, scalable technology solutions. Its multidisciplinary teams bring together expertise in artificial intelligence, software engineering, digital product design, data, business intelligence and robotics to solve complex business and operational challenges. From opportunity discovery and rapid prototyping to deployment and scaling, Osedea combines technical excellence, human-centred design, and a pragmatic focus on measurable value to build technologies that perform in the real world and create lasting impact.
From exploring possibilities and developing prototypes to deployment and scaling, Osedea works closely with its clients at every stage. Its approach combines technical excellence, pragmatism and a deep understanding of human needs to design solutions that perform in the real world and generate lasting impact.
