Deloitte unveils insights on the accelerating adoption of physical AI and robotics

Physical AI: Bridging digital intelligence and the physical world

Physical AI describes AI systems that help machines perceive, understand, reason, and act in the physical world -continuously and in real time. This includes a new generation of robots, autonomous vehicles, and advanced sensor platforms. Unlike older robots that simply repeated instructions, physical AI systems can learn from experience and adapt their behavior as conditions change. In simple terms: the software is becoming much better at “understanding” the real world, not just following a script.

Several technologies are driving this fast progress:

• Multimodal Vision-Language-Action (VLA) models, which help robots interpret what they see, connect it to language or instructions, and choose the next action.
• Onboard neural processing units (NPUs), which run AI directly on the machine with very low delay, reducing reliance on cloud computing.
• Stronger robotics hardware, including improved computer vision, better sensors, and more capable actuators (the components that create movement and force).

Together, these advances—combined with improving cost economics—are pushing physical AI into broader industrial use.

“The physical world is inherently dynamic. The challenge lies in simulating these variations so robots can learn to adapt and interact with uncertainty, similar to humans. We also face hardware limitations, as conventional robots often can’t lift half their weight due to actuator limitations, unlike human muscles. Additionally, real-time processing is crucial for safety-critical decisions, where even a one- or two-second delay can have severe consequences.”

Overcoming barriers and expanding impact

Physical AI has major potential, but scaling it is not simple due to several key challenges:

• The “reality gap”: robots trained in simulations can still perform differently in the real world.
• Trust, safety, and reliability: even small mistakes can cause damage or injuries when machines operate physically.
• Regulation and compliance: rules differ by country and industry, and they are evolving quickly.
• Data complexity: physical AI often relies on large volumes of mixed data (video, audio, sensor readings, location signals).
• Human acceptance: organizations must address concerns about jobs and focus on human–machine collaboration, not replacement.

Even with these barriers, Deloitte observes that physical AI is spreading beyond the usual industrial settings:

• Healthcare: AI-enabled robotic surgery and autonomous imaging tools can support staff shortages and improve precision.
• Restaurants: robots are being used for food prep, kitchen support, and serving, helping reduce labor pressure.
• Utilities: companies such as Naturgy Energy Group are using drones for grid inspection and expect robots to take on more dangerous field tasks to help prevent injuries and fatalities.
• Public services: AI-powered drones support infrastructure inspections, and autonomous shuttles can improve mobility and accessibility.

Humanoid robots and what comes next

Humanoid robots are presented as the next major step. Their human-like shape and dexterity matter for a practical reason: they can move through spaces designed for people-factories, warehouses, offices, homes—without expensive redesigning. As AI reasoning and “agentic” systems improve, these robots are becoming better at planning multi-step tasks, adjusting to unfamiliar situations, and recovering when something goes wrong.

Analysts forecast strong growth in humanoid deployments over the next decade, with long-term projections that extend into the trillions in market value by 2050. Deloitte notes that warehousing and logistics are emerging as early testing grounds, driven by labor shortages and the need for careful, precise handling of objects.

Beyond humanoids, researchers are exploring more experimental directions, including biologically integrated machines and early ideas in quantum robotics, which promise unprecedented speeds and capabilities, although useful quantum robots are still decades away.

“We aimed to build a human-centric, multipurpose robot that could move like animals or people and thrive in human spaces. The most important thing is that each of our robot's features has a clear purpose. We are capturing the function that underlies that form. For example, lifting and moving bins requires a narrow footprint, dynamic stability, and bimanual capability, which a bipedal, upright humanoid form provides most effectively.” 

The journey of physical AI, from autonomous systems inspecting power grids to humanoids assisting in rehabilitation centers, signifies a fundamental shift in how we conceive and interact with machines. These breakthrough technologies are moving beyond simple task automation, ushering in an era of entirely new categories of adaptive and intelligent systems that will redefine the future of work and daily life.

Read the original article on the Deloitte Insights website