Closing the Gap Between Advanced Robotics and AI Research for Real-World Deployment

Tell us about yourself and your role at Intrinsic.

I am the Chief Commercial and Product Officer at Intrinsic, an AI robotics group at Google. I lead our commercial and product strategy, helping turn advances in AI and robotics into practical tools for industrial automation.

I spent fifteen years in technical and leadership roles at IBM’s Software Group and Research Division before moving into robotics. I served as Executive Director of Solutions at Willow Garage and later as CEO of Redwood Robotics, which was acquired by Google. At Google, I led product efforts across Cloud AI, robotics and search infrastructure. I then became Chief Product Officer at Fetch Robotics, overseeing product strategy for autonomous mobile robots used in logistics and warehousing.

Across these roles, my focus has been on making intelligent software useful in the physical world: helping businesses build robotic systems that can adapt to their needs and deliver value over time.

What is the manufacturing problem you are most focused on solving?

We want to help manufacturers automate more of the work that has traditionally been too difficult or expensive to address, particularly where products vary, batches are small and processes change frequently.

One major problem is the amount of custom automation engineering required for each solution. Before developers and integrators can focus on the manufacturing task, they often have to configure hardware, calibrate sensors and connect the software for perception, planning and control. Too much of that effort is difficult to reuse.

Intrinsic Core addresses this challenge by making foundational components of our platform available as open-source software that developers can run locally. It gives them a starting point for building and testing applications, so they can spend more time solving the manufacturing problem itself.

For manufacturers, the opportunity is to make automation practical across more tasks and respond more quickly when production needs change.

Industrial robots have been used for decades. Why does deploying and adapting automation remain so difficult for many manufacturers?

Industrial robotics has been particularly successful in environments where the same process is repeated at high volumes. The engineering investment can be spread across a large number of identical operations.

That becomes harder when a manufacturer handles many different products in smaller batches. A change in a part’s shape, position or material can require adjustments to the tooling, sensing and robot behavior. By the time a system is ready, the production requirements may have changed again.

The difficulty also extends beyond the robot. Cameras, grippers, controllers and application software all need to work together. When those connections are built specifically for one workcell, adapting them to another task or a different product to be manufactured can mean significant rework.

Reusable software components and more consistent interfaces can help developers carry more of that work forward. The aim is to make each new application an extension of what they have already built, with less engineering starting from zero.

 

There is enormous excitement around AI and robotics today. Where is AI already delivering meaningful value on the factory floor, and where is the hype ahead of reality?

Some of the clearest applications are in computer vision and inspection, where AI helps systems recognize parts or identify defects. In robotic manipulation, AI can help determine a part’s position, orientation and how to grasp it. These capabilities are useful when parts do not arrive in exactly the same place or condition each time.

Machine-tending is a practical example. Our Open Machine Tending Solution brings together simulation, perception-driven pose estimation, motion planning, and grasp planning in an open reference design that developers can run locally and customize.

Where the hype gets ahead of reality is the assumption that a successful demonstration means a system is ready for sustained production. A factory application must handle variation, recover from failures and meet requirements for safety, quality and cycle time.

Simulation and local testing help developers evaluate applications earlier. Dependable deployment still requires testing against the conditions and exceptions the system will encounter on the factory floor.

How is Intrinsic helping close the gap between advanced robotics and AI research and dependable, real-world deployment?

We are giving developers access to foundational technology, practical examples and a path to build on that work.

Intrinsic Core brings together open-source components for capabilities such as motion and grasp planning, pose estimation and real-time control. Developers can use these locally to build and test applications, integrate their own assets and work with existing ROS software.

The purpose is to address some of the foundational work that slows robotics development down: setting up an environment, connecting modular capabilities and building basic motion and control workflows. Developers can use standardized models and reusable components, while still importing their own assets and extending applications for specific tasks.

We are also providing concrete starting points. The Open Machine Tending Solution shows how these capabilities work together to load and unload parts from a CNC machine. Top teams in our AI for Industry Challenge used components we are now open-sourcing to tackle complex cable handling and insertion in electronics assembly. Their work illustrates what developers can build with these capabilities, while production deployment requires further validation.

For developers moving toward development and scale, there is an optional path to Intrinsic Enterprise services. Applications built on Intrinsic Core share architectural models with other Intrinsic services, which can help reduce the amount of refactoring required when a developer moves from testing toward deployment, operation and scale. The open-source environment is not intended to replace the engineering required for production; it is intended to make the path toward production more accessible and efficient.

This initial release is also the beginning of an ongoing effort. We are committed to expanding the available capabilities, improving ROS compatibility and using community feedback to guide development.

Factories are dynamic: parts vary, layouts change and production runs are getting shorter. What capabilities are needed for robots to adapt without constant reprogramming?

Robots need to sense changes and adjust their actions within clearly defined operating limits. That means recognizing parts and their positions, planning appropriate motions and using force feedback during execution.

The software also needs to be adaptable. If an application is tightly tied to one part or workcell layout, even a small change can require substantial rework. Reusable “skills” and consistent interfaces make it easier to update the relevant parts of an application.

For example, developers can customize the Open Machine Tending Solution by importing their own CAD parts and extending the application logic. That provides a baseline for testing changes instead of rebuilding the entire application.

The goal is to reduce how much developers must manually specify for every variation. Reliable control, accurate sensing and testing remain essential to making that flexibility useful in production.

What role must robot manufacturers, system integrators and software developers each play in making flexible automation easier to deploy at scale?

Robot manufacturers can make hardware easier to integrate through well-supported drivers, consistent interfaces and clear documentation.

System integrators bring the manufacturing expertise needed to turn those capabilities into working applications. They understand the process, safety requirements and operational constraints. Reusable software can help them spend more time configuring and optimizing solutions for customers and less time recreating foundational capabilities. Transitioning into an asset-based business model with a more attractive cost structure and higher margins presents a significant opportunity for them.

Software developers can provide those reusable building blocks and the tools to connect, simulate and test them. They also have an opportunity to share improvements that benefit applications beyond their own.

Intrinsic Core is designed to complement ROS and connect with existing workflows. Our responsibility goes beyond releasing code: we are continuing to improve compatibility, make the components more modular and engage developers in shaping what comes next. Making flexible automation easier will require that kind of sustained collaboration across the ecosystem.

How will the jobs of robotics engineers, integrators and factory operators change as AI-powered and low-code tools become more widely available?

These tools can make expertise go further by reducing repetitive development work. Robotics engineers may spend more time designing application behavior, testing difficult cases and improving reliability. Integrators may be able to apply more of their experience across multiple customers as software and workcell designs become easier to reuse.

Factory operators and manufacturing engineers remain essential. They understand how the process actually behaves, including the material differences and everyday exceptions that a development team may not anticipate. Their input is critical to deciding what to automate and evaluating whether it works.

As the tools become more accessible, more people can contribute to building and improving applications. Understanding the manufacturing process, evaluating system behavior and troubleshooting failures will remain valuable skills. Leveraging AI-powered interfaces to build a knowledge base of manufacturing processes and best practices is a key opportunity for the next cohort of operators and process engineers.

Many manufacturers are shifting toward modular production lines to keep up with changing product cycles. How does combining modular hardware with AI-driven software change the economics of setting up a new factory floor?

The economic benefit comes from being able to reuse more of the investment when production needs change. Modular hardware makes a workcell easier to reconfigure, while adaptable software can reduce the engineering needed to handle new parts or tasks.

That matters because the purchase price of a robot is only part of the cost. Integration, programming, validation and downtime during changeovers also affect whether an application makes financial sense.

If a manufacturer can adapt an existing workcell and preserve useful application logic, it may be able to spread those costs across more products. AI can help the system handle variation, while reusable software provides a foundation for testing and implementing changes.

The savings will depend on the application and the work required to integrate and validate it. But lowering the cost of change can make automation practical for shorter production runs and a wider range of tasks, giving manufacturers more options as customer demand evolves.

The content & opinions in this article are the author’s and do not necessarily represent the views of RoboticsTomorrow

Featured Product

ResinDek TRIGARD ESD ULTRA FOR HIGH-TRAFFIC ROBOTIC APPLICATIONS

To maximize the productivity of an autonomous mobile robot (AMR) or automatic guided vehicle (AGV) deployment, it’s critical to create the optimal environment that allows the vehicles to perform at their peak. For that reason, Cornerstone Specialty Wood Products, LLC (www.resindek.com) created the TriGard ESD Ultra finish for its ResinDek engineered flooring panels. The TriGard ESD Ultra finish is ideal for high-traffic robotic applications characterized by highly repetitive movement patterns and defined travel paths.