In an AI future, Toyota Research Institute still sees room for human touch

LOS ALTOS, Calif. — A mock grocery store inside a Toyota Research Institute laboratory here presents challenges for a humanoid robot plying the aisles.

The robot must know where items on its shopping list are located and how to navigate within reach. It must understand how much each product weighs and how it's packaged.

Should the robot use pincers affixed to its right arm to grab an object — such as a milk jug — or the vacuum-based suction tube on its left hand to pluck a product from the shelf and place it in the cart?

There's a fervor around the burgeoning robotics industry, which Boston Consulting Group projects will rise from about $40 billion in annual market sales now to a $160 billion to $260 billion range by 2030. But such bullish outlooks remind Max Bajracharya, senior vice president of robotics at the institute, of another much-ballyhooed technology.

"It is robotaxis a decade ago," he said. "What they didn't account for then was we didn't have the technology to achieve that, and we're seeing that again."

What's missing, in his view, is an underlying set of technologies that make robots both useful and reliable in the real world. Bajracharya's job is not necessarily to develop an end product such as a grocery shopping robot, but to discover and explore the bounds of those foundational technologies.

That's a mindset that permeates beyond robotics and into all corners at the Toyota Research Institute. Established as the research arm of its namesake automaker, the institute concentrates on exploring five fields: energy and materials, human-centered artificial intelligence, machine learning, human-interactive driving, and robotics.

Lately, the organization has advanced generative artificial intelligence. Researchers showcased a new AI technique in June that incorporates engineering parameters earlier in the car design process, thereby shortening design cycles.

On Tuesday, researchers disclosed a "breakthrough" generative AI approach that teaches robots more than 60 skills, including pouring liquids and using tools, without writing a line of code. The institute expects that will help its robots learn hundreds of skills by the end of the year.

It's a landmark step toward making robots more versatile in a wider range of environments.

"If we've done our job right, these can apply to many different applications," Bajracharya said.

Those AI-related developments belie a more old-fashioned organizational goal — using new technologies and tools to improve the human condition, the company said.

Gill Pratt, CEO of the institute and chief scientist for Toyota Motor Corp., acknowledges that sounds like marketing pablum. Yet it's central to the organization's ethos: Technology should support humans, not supplant them.

While there are broad concerns that AI's growing sophistication will ensure human obsolescence, Pratt offered a reminder the future is not yet scripted.

"We have a choice," he said. "What we do at TRI is very much always try to figure out how to use this stuff in service of human beings by starting with the human beings first."

For example, when other automakers pursued the highest reaches of self-driving technology and advanced driver-assistance systems that lessened the role of humans, Toyota went in the opposite direction. It instead developed driver-assistance software it calls Guardian. Using cameras, Guardian watches the road but lets humans retain their usual role controlling the vehicle. Guardian runs in the background and only intervenes if it detects an imminent hazard.

Rather than asking humans to oversee an active automated system, Toyota's Guardian uses automation to oversee and support an engaged human driver.

The roots of this collaborative strategy are found in a prescient 1960 paper titled "Man-Computer Symbiosis" by psychologist and computer scientist J.C.R. Licklider. Rather than buy the premise of AI that eventually renders humans obsolete, Licklider inverted that and supported the notion of IA — intelligence amplification.

He studied and implemented that approach at both the Massachusetts Institute of Technology and the Defense Advanced Research Projects Agency. Decades later, Pratt followed in his footsteps at both organizations and adopted those principles. He then brought intelligence amplification ideas to the institute, where he's worked for the past seven years.

"It's an incredibly brilliant idea," Pratt said. "It's like, 'Let's build machines to amplify human beings and not replace them,' " Pratt said. "And that's been the basis of everything we have done."

Whereas Guardian involves specific safety-critical situations, the institute's human-interactive driving division is exploring ways to incorporate AI into the human driving experience.

Researchers are culling data from tests in both simulators and on racetracks to try to understand how motorists' awareness levels correlate to steering, braking and throttle inputs. They're further monitoring how drivers behave when they encounter yellow lights.

By analyzing and predicting responses, they hope to blend assistive features into the driving experience in a way that feels intuitive and sensible to humans. They further hope they can use AI as a driving coach of sort. It might help humans form better individual habits and incentivize good behavior.

"Imagine a car that you put your teenage children in, and you don't have to be there," said Avinash Balachandran, senior director of the human-interactive driving division. "You're confident the car is going to make them a better driver."

The notion that R&D efforts should still orbit around the idea of a human driver runs counter to many industry sentiments. And it's not the only area in which Toyota has flashed a contrarian streak.

The automaker has said it would spend $70 billion on electrification efforts. It has committed to ensuring a third of its approximately 10 million vehicles sold globally each year are EVs by 2030. But Toyota says reducing carbon emissions should be its primary goal, not selling electric vehicles.

Toyota believes there are alternatives, namely hydrogen fuel cells, that can overcome battery-electric limitations in specific applications, such as long-haul trucking, and achieve a faster path toward net-zero emissions.

"We believe we need diverse solutions," said Brian Storey, senior director of the institute's energy and materials division.

It is his job to find and enhance them.

The division develops new materials for batteries and fuel cells as well as hydrogen electrolyzer and solar technology. It's working on battery chemistries that need less rare-earth material and researching solid-state battery technology. The institute works with several university partners and the National Renewable Energy Laboratory. Part of their work utilizes new algorithms that make faster sense of datasets than humans could understand on their own.

It can take almost 50 years for new chemistries to transform into mature products, Storey said, and "a lot has to happen in a short amount of time" for the industry to meet carbon-reduction targets.

Part of his mission is to shrink that time frame by utilizing new machine-learning techniques. It helps focus research when "there's an infinite number of possibilities, and we've only scratched the surface in the kinds of things we can make," he said.