We still seem so far from the “Jetsons robot” – where a bipedal “bucket of bolts” is doing things like cleaning our kitchens and folding our laundry – but there’s ample evidence that this is soon about to change.
With the combination of computer vision and sophisticated neural nets, robots can now “think” much like we can, in addition to being able to “see” much like we can as well. That is likely to lead to a robot revolution the likes of which we cannot imagine.
Speaking of Jetsons, I wrote recently about the Nvidia Jetson architecture, which, it turns out, is a candidate for a robot “brain” that can help some mechanism accomplish a lot.
“The powerful edge processing offered by Jetson Thor will take Digit to the next level — enhancing its real-time responsiveness and expanding its abilities to a broader, more complex set of skills.” said Peggy Johnson, CEO of Agility Robotics, in an Nvidia blog on the system’s release. She was referring to a humanoid robot developed by her company specifically for material‐handling and logistics tasks: for example, picking up totes, unloading bins, moving packages in warehouse/distribution environments. Standing about 5′9″ (approx. human waist height) and with a payload around ~35 lbs., Digit has bipedal locomotion, allowing it to navigate environments built for humans (stairs, platforms, narrow aisles) rather than wheels alone. So theoretically, it can get into our environments, and mimic some of the tasks that we do.
Enterprise Impact
It’s just a hop, skip and a jump from here to widespread workplace adoption of robotic workers.
A recent panel segment at Stanford addressed the likely outcomes for business. Ayush Khandelwal spoke with Tara Javidi, Jacob Rodriguez and Michael Chen about the rise of robotics in the age of AI.
We try to reduce it to the environmental engineering problem,” said Rodriguez, whose company, Oligo, works on space installations and equipment. “We’re working on concepts like generative design schemes, to produce topology-optimized structures. We’re doing our own harnessing trying to select systems requirements. So a lot of our work on the research side is (related to) ‘how do the hardware engineers get all the requirements to build the robot?’”
He cited a “sim to real gap” that requires more research.
“A lot of our work is, ‘how do we actually create test-as-you-fly conditions on the ground to (ensure the right result)?’” he said.
Chen had some profound input on the simplification of design to provide real results.
“Instead of focusing on having a robot that can mow your lawn, and do your laundry, and wash your dishes, we’re focusing on one task,” he said. “What that allows us to do is we can actually saturate the whole possible environment of all the different situations that robot would potentially encounter while doing a particular task in industry, and by simplifying that and focusing on less experimental ways of collecting data, by sticking to real world data and human demonstrations, we’re able to actually bring something that can do real work, in the real world, not a decade from now, but real, useful humanoid robots today.”
Javidi pointed out some obstacles to enterprise planning in this space, including both siloed data and collaborative processes.
“The first (challenge) in bringing AI into the physical world has been social,” she said, “in the sense that you’re building an engineering system, but you rely on a lot of software and AI, so kind of bringing the team to talk to each other (about) the complexities that are caused by building the software and so on (is important).”
Solve My Pain
Another big topic that came up is the need to target applications to customer needs.
“I have an AI group that wants to build systems that can generalize for all physical spaces, and then we have customers who don’t care if we are building the new cool thing that will translate to some other space,” Javidi said. “They want to solve their problem … you work with what the customer’s biggest pain is.”
At the same time, she noted, it’s important to avoid problems like overfitting, and take a practical approach to a given solution.
Rodriguez agreed, talking about building out systems, and what kinds of thinking apply.
“What does a long range, mid-range, short range planner look like?” he asked rhetorically. “You basically have different definitive profiles for the hardware you work with.”
He also stressed the value of having more general or broader software run specialized systems.
“We have to create the system to build the hardware specific to the application, and let the software stay general,” he said. “I try to containerize the environmental conditions to the exact use case, but our software stack, even for spacecraft, stays generally the same.”
Chen talked about the practicality of deploying systems initially in more controlled environments. That includes “dark runs” where the technology can explore without a lot of human intrusion.
“The most important thing is demonstrating (something) working very well around no people at all first,” he said. “So when we do our initial deployments, it’s dark. You guys have probably heard of ‘dark warehouses,’ where there (are) no people around … we’re able to deploy (the technology) in a real customer facility and create value without actually (having) to worry about (people getting hurt) potentially, because a robot falls or hits someone by accident, and then having the right safety precautions in place, so that people don’t get too close to the robots and get smacked by accident.”
The Jobs Question
Later in the talk, there was this interesting part about the potential for job displacement. I think Chen in particular brought some good insights to the eternal question of: will robots augment or replace human workers? He urged stakeholders not to jump to conclusions about robots replacing people.
“In many cases, with the customers we’re working with, what’s actually happening is reallocation,” he said, presenting a dystopian picture of a given pre-robotically-integrated work environment. “So warehouse managers, they are ‘hair on fire,’ constantly running around. Their workers are constantly overworked and stressed, what they’re missing is being able to actually tackle the problems that allow their businesses to grow. They’re so focused on just the most basic tasks just to keep the business alive.”
With the right integration, he suggested, companies will be able to create synergy, to create efficiency, so that both the workers and the business can thrive – but that, again, comes with addressing those pain points.
“What we’re hoping,” Chen added, “is that our technology enables warehouses and warehouse owners to be able to reallocate that labor towards things that are customer-facing, towards applications that require the skills humans are really good at, like context switching, like problem solving, like creativity. And because of that, it implies that we need to have really good human-robot interaction for that to work.”
If you have gotten this far, I think you have some good ideas around what’s happening with robotics, in enterprise. You can watch the video, too, and engage with other aspects of what these experts and others have to say about what our workplaces will look like in 2026, in 2027, and beyond. Because they probably won’t look the same.
