Physical AI and robotics are moving beyond impressive demonstrations into a new phase of practical deployment, with companies now targeting specific, high-value use cases in manufacturing and logistics — production-ready systems capable of delivering measurable ROI.

After years of research breakthroughs and impressive demonstrations, the focus for physical AI has shifted to real-world data, functional safety and sustainable business models. The next phase will be defined by practical deployment, from humanoid robots navigating factory floors to cognitive systems that can see, hear, feel and adapt to the environments around them.

“It’s a huge issue when you think about labor, when we need it to produce and produce at scale,” said Andrew Lonsberry (pictured, right), co-founder and chief executive officer of Path Robotics Inc. “We need products, and we need them built at scale now. We need robotics that are intelligent enough to go do those jobs. That’s what we built Path to do.”

Lonsberry and other industry experts spoke to John Furrier, executive analyst at theCUBE Research, during the Machina AI Summit, the exclusive day-zero event preceding the RAISE Summit in Paris. The interviews highlighted the opportunities, challenges and hard lessons emerging as organizations move beyond pilots into production — and where physical AI and robotics are heading next. (* Disclosure below.)

Here are 10 standout insights from the event:

1. Path Robotics cuts welding labor from 150 hours to 9 with AI-powered robots.

Path Robotics is solving the severe skilled-labor shortage in U.S. manufacturing, where the average welder is 55 years old and 20–30% are expected to retire soon. The company augments human workers with AI-powered robots that use a neural network called Obsidian and a real-world data flywheel to continuously improve — getting better with every deployment rather than degrading like traditional robots, Lonsberry said. In one customer case, human welding time on truck chassis and data-center skids dropped from 150 man-hours to just 9, a 91% reduction that enables manufacturers to scale production amid surging demand.

Check out theCUBE’s complete interview.

2. VicOne secures physical AI and robotics with full-lifecycle protection.

VicOne Inc. provides comprehensive security for physical AI systems in robotics and automotive across the full development lifecycle, including software bill of materials management, vulnerability handling, regulatory certifications such as CRA and ISO 21434, in-device protection and continuous robotic monitoring. The company is helping manufacturers secure supply chains by detecting component origins and mitigating AI-specific risks that can fool robots into unsafe behavior, explained Jason Yueh, chief technology officer of VicOne. As physical AI moves from demonstrations to production deployments in open environments, cybersecurity is the foundation of safety, he added.

Watch the full interview from theCUBE.

3. Boston Dynamics accelerates humanoid robotics commercialization.

Boston Dynamics Inc. is commercializing its electric Atlas humanoid after years as a research platform, capitalizing on the explosion of demand for humanoids paired with rapid AI advances. AI has dramatically accelerated behavior development, reducing the time to move new capabilities from simulation to robot from roughly a year to just a few hours with near-99% reliability, Amanda McMaster, CEO of Boston Dynamics, highlighted. The company is deploying Spot for industrial inspection in manufacturing and food and beverage, while bringing Atlas into net-new use cases such as part sequencing and machine tending in auto manufacturing.

See theCUBE’s full coverage.

4. Skild AI builds a universal brain for heterogeneous robot fleets.

Skild AI Inc. is developing the Skild Brain — a single universal AI model that powers multiple robot form factors, from humanoids and quadrupeds to traditional arms, enabling them to share learnings and adapt across tasks and environments. The company pre-trains on large-scale video and simulation data before refining with real-world teleoperation, allowing robots to handle uncertainty in unstructured settings, explained Abhinav Gupta, co-founder and president of Skild AI. This omni-bodied approach recently demonstrated powerful in-context learning when a robot adapted on the fly to a broken ankle motor it had never been trained on.

Watch theCUBE’s full sit-down.

5. Apptronik advances humanoid robotics with Apollo and Robot Park.

Apptronik Inc. is commercializing its Apollo humanoid robot after a decade of development and recently launched Apollo 2 alongside Robot Park, a 90,000-square-foot data factory collecting multimodal training data from teleoperation, human video and synthetic sources. The company is focused on back-of-house applications in manufacturing and logistics, running pilots with major customers while advancing dexterous manipulation and modular designs that support both wheeled and legged mobility, according to Jeff Cardenas, co-founder and CEO of Apptronik. The current moment mirrors the early PC era — real technological progress amid hype — with robots positioned as foundational infrastructure, he noted.

Check out the full story on theCUBE.

6. Agility Robotics brings functionally safe humanoids into human workspaces with Digit.

Agility Robotics Inc. is commercializing Digit, a human-centric, multipurpose humanoid robot designed to safely operate in unmodified human environments. The upcoming Digit V5 introduces functional safety features — including certified safe electronics, human detection and controlled fall behavior — enabling the robot to step out of cages and work cells for the first time at scale, said Jonathan Hurst, co-founder and chief robot officer of Agility Robotics. Early high-value use cases center on warehouse workflows such as bin picking and de-palletizing in narrow spaces, where the humanoid form factor delivers clear advantages over specialized automation.

Don’t miss theCUBE’s full interview.

7. Niantic Spatial provides real-world spatial models for physical AI.

Niantic Spatial Inc. is building real-world foundation models and visual navigation systems that give robots and autonomous systems high-fidelity 3D spatial awareness for operating in complex indoor-outdoor and GPS-denied environments. The company creates detailed 3D reconstructions from low-cost cameras, converts Gaussian splats into aligned meshes for efficient deployment and is advancing semantic understanding, according to Inhi Cho Suh, CEO of Niantic Spatial. By grounding models in real-world data rather than simulation alone, the company helps robots navigate messy, dynamic spaces more safely and effectively, addressing one of the biggest barriers to scaling physical AI, she emphasized.

Catch the full story on theCUBE.

8. Rhoda AI enables intelligent robot manipulation with Direct Video-Action.

Rhoda AI Corp. is building a foundation model that trains robots on internet-scale video data for pre-training and robot actions for post-training, enabling adaptive manipulation in variable real-world conditions rather than rigid pre-programmed motions. The company’s Direct Video-Action approach lets robots generate videos of tasks in real time and convert them into precise actions, addressing the data bottleneck that has limited most deployed robots to repetitive tasks, Andrew Wooten, co-founder and chief product officer of Rhoda AI, pointed out. With a recent $450 million Series A and a team of nearly 100, Rhoda is moving from successful factory pilots in automotive to its first commercial deployments in manufacturing and logistics later this year.

Watch theCUBE’s complete interview.

9. Neura Robotics builds cognitive robots with multimodal sensing.

Neura Robotics GmbH is developing cognitive robots that perceive and interact with the physical world through multimodal sensing and autonomous decision-making in dynamic environments. The company takes a full-stack approach — hardware, nervous-system-like reflexive compute and the Neuraverse platform — because existing hardware and architectures were insufficient to deliver the right sensor data and millisecond-level reactions required for real physical AI, David Reger, founder and CEO of Neura Robotics, explained. By defining standards and then opening key layers while enabling companies to train and own their proprietary knowledge via “brain accounts,” Neura Robotics aims to accelerate scalable, safe deployment amid massive global labor shortages, he added.

Don’t miss the full segment on theCUBE.

10. SemiAnalysis sees robotics shifting to practical, application-specific deployments.

SemiAnalysis LLC observes that the physical AI and robotics industry is moving from broad general models toward practical, application-specific use cases where robots can already deliver value in defined warehouse and factory “pockets.” Pilots are converting into paying customers and repeat business, with companies using sequenced tasks to build revenue and data flywheels rather than chasing immediate generality, explained Reyk Knuhtsen, analyst at SemiAnalysis. China’s massive scale advantage in humanoid production and supply chain strength points toward 2027 as a breakout year for real commercial deployments, even without a single ChatGPT-style moment, he said.

Explore the full segment on theCUBE.

Here’s the complete video playlist from SiliconANGLE’s and theCUBE’s coverage of the Machina AI Summit:

(* Disclosure: TheCUBE is a paid media partner for the Machina AI summit. Sponsors do not have editorial control over content on theCUBE or SiliconANGLE.)

Photo: SiliconANGLE