Claire chatted to Aaron Edsinger from Hello Robot about their Stretch robots that can help people with mobility issues live more independently.
Aaron Edsinger is CEO and Co-Founder of Hello Robot, a robotics company dedicated to developing practical, safe, and accessible robots for people. With a passion for building robots and robot companies, Aaron has founded four companies focused on bringing human collaborative robots to the consumer, manufacturing, healthcare and research markets. As Robotics Director at Google, he led the business, product and technical development of two of Google’s central investments in robotics, and helped to define the overall robotics strategy.
Deborah Lupton / Pop Chips / Licenced by CC-BY 4.0
Intrinsic (an AI robotics group at Google) announced that the company is making parts of its platform open source. Intrinsic Core
is a set of ROS-compatible capabilities for building sophisticated robotic applications.
The capabilities included in the open source package are as follow:
Alongside this package, Intrinsic has also released Open Machine Tending Solution (OMTS). This is a ready-made reference design for a real-world use case. The intention is to give developers a starting point to customize their own solution, without needing to build from scratch.
You can download Intrinsic Core on GitHub here.
Your forklift breaks down at 7 a.m. Your best technician is 40 minutes away. Nobody has even checked whether the part is in the warehouse yet. By the time someone finds the manual, confirms the part, and gets a dispatcher moving, half a shift is gone and the SLA clock is still running.
Nobody’s building a demo around that.
Every agentic AI pitch in your inbox right now is about something else: an agent that drafts your sourcing emails, summarizes supplier contracts, or writes better product copy. All useful. But none of them help when the oven in your industrial kitchen goes down, the excavator on your job site won’t start, or the truck carrying a customer’s order breaks down on the interstate.
That’s the gap. Some of the most valuable AI use cases in manufacturing aren’t particularly glamorous, and they’re easy to overlook.
More than six in ten manufacturers experienced unplanned downtime in the past year, costing the sector as much as $852 million a week. At an average cost of $1.7 million an hour, a single incident can run as high as $42.6 million, according to a 2025 global survey from Fluke Corporation and Censuswide of 600 manufacturers in the US, UK, and Germany.
And it’s not for lack of trying.
L2L’s October 2025 survey of more than 600 manufacturing leaders found that 60% lose more than $250,000 a year to preventable downtime. Facilities average 30 hours of downtime a month, with more than half of it unplanned. And while 93% of respondents said they’d taken action to address it, only 46% saw a measurable improvement.
That’s a pretty big gap between the effort going in and the results coming out.
The difference can be significant. In Aquant’s 2025 Industrial Machinery Service Benchmark Report, leaders fix problems on the first visit 87% of the time and resolve them in an average of two days. Bottom performers take eight days, at nearly eight times the cost per event. The report points to agentic AI built for the complexity of industrial service as one factor separating the two groups. The alternative is still a lot of calls, handoffs, and people trying to find the right information.
There’s a reason the flashy use cases get the demos while the unglamorous ones carry a much bigger bill.
Sourcing emails and contract summaries are relatively self-contained. Asset uptime isn’t.
Fixing an equipment problem might mean:
That’s a harder problem to solve. It’s also where the financial impact can be much bigger.
One critical piece of equipment sitting idle for an extended period can become a million-dollar-a-day problem at Fortune 500 scale. Across those companies, that can add up to a trillion-dollar issue. Solve it, and the payoff is tangible: fewer trucks making repeat trips, less capital tied up in idle equipment, and fewer SLAs quietly slipping.
DataRobot has built this kind of approach for manufacturing and industrial customers. Instead of asking one generalist agent to do everything, the work is broken into agents that each handle a specific part of the workflow:
Working together, those agents have increased first-time fix rates by 50% to 60% for customers running the solution. That’s the kind of improvement that can move a service organization from spending days resolving an issue to getting the equipment back up and running much faster.
Marc Amarillas, Director of Customer Success Engineering at DataRobot, walked through this setup on a recent Supply Chain Now webinar with host Scott Luton, co-host Kim Reuter, and me. The conversation focused on what it takes to move agentic AI from pilot to enterprise scale.
His point was simple: an asset going down can be a “million-dollar-a-day problem” at a Fortune 500 company. Across the number of critical assets those companies operate, that becomes a much bigger issue.
Physical assets are the clearest example, but they’re not the only one. The same math shows up anywhere a manufacturer’s money is exposed while people are trying to figure out what to do next.
On the same webinar, I shared how a tariff and labor signal put one global manufacturer at $2.5 billion in revenue at risk, essentially overnight.
The old process was slow: size up the exposure, check every supplier tier, model possible responses, and decide what to do. That could take weeks.
DataRobot built five agents for the customer, integrated with SAP, to handle that work. The agents detected the signal, quantified the exposure, modeled three responses, and helped reallocate resources globally, with a human approving every step.
The result:
The use case is different from asset management, but the approach is the same: find the expensive problem, break the workflow into manageable pieces, and keep a person in the loop when a decision is difficult to reverse.
None of this matters if the technology can’t run where a manufacturer’s data already lives.
Most manufacturers have infrastructure commitments they can’t simply walk away from: SAP, specific cloud environments, and on-premises systems holding sensitive supplier and customer data. An agent that requires everything to move to a new environment before it can help isn’t going to get very far.
That’s why DataRobot supports the same deployment and governance approach across managed SaaS, private cloud, self-managed, and fully air-gapped environments, while integrating with the SAP, ERP, and on-premises systems manufacturers already use.
As I put it during the webinar, the AI should come to the supply chain owner wherever they need it to run, not the other way around.
That’s an important difference between an agent that stays a demo on someone’s laptop and one that an IT and security team will actually allow to work with production data.
You don’t have to bet the whole operation on AI at once.
Start with the workflow where the cost is already visible on the P&L, whether that’s an idle asset, exposed revenue, or another problem that’s taking too much time and money to resolve.
Build an agent workforce around that one problem first. Deploy it inside the systems you already have. And build the governance and audit trail in from day one, rather than trying to add them later.
The chatbot use cases will keep getting the headlines.
But the forklift, the oven, or the tariff signal nobody caught until Monday morning? That’s where a lot of the real money is sitting.
Start there.
See how DataRobot can help you build and deploy agentic AI around the supply chain workflows that matter most to your business.
The post The unsexy AI: why your forklift matters more than your chatbot appeared first on DataRobot.
The 2026 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2026) will be held from 27 September – 1 October in Pittsburgh, USA. The programme includes plenary and keynote talks, workshops, tutorials, forums, competitions, and a debate.
The keynote talks will be held from 28 – 30 September:
Community discussion sessions open to all attendees, these forums will be held on Monday 28 and Tuesday 29 September.
Three plenary panels will be held:
There are 44 workshops and tutorials planned for Sunday 27 September, with a further 42 scheduled for Thursday 1 October. You can find out more about these here.
Join Sabine Hauert (University of Bristol and Robohub), Ella Scallan (Robohub), and Evan Ackerman (IEEE Spectrum) for a science communication crash course. This will take place at 11:00 on Tuesday 29 September. Find out more here.