Engineers at Rice University have developed a method that allows humans to help robots “see” their environments and carry out tasks.

The strategy called Bayesian Learning IN the Dark — BLIND, for short — is a novel solution to the long-standing problem of motion planning for robots that work in environments where not everything is clearly visible all the time.

The peer-reviewed study led by computer scientists Lydia Kavraki and Vaibhav Unhelkar and co-lead authors Carlos Quintero-Peña and Constantinos Chamzas of Rice’s George R. Brown School of Engineering was presented at the Institute of Electrical and Electronics Engineers’ International Conference on Robotics and Automation in late May.

The algorithm developed primarily by Quintero-Peña and Chamzas, both graduate students working with Kavraki, keeps a human in the loop to “augment robot perception and, importantly, prevent the execution of unsafe motion,” according to the study.

To do so, they combined Bayesian inverse reinforcement learning (by which a system learns from continually updated information and experience) with established motion planning techniques to assist robots that have “high degrees of freedom” — that is, a lot of moving parts.

To test BLIND, the Rice lab directed a Fetch robot, an articulated arm with seven joints, to grab a small cylinder from a table and move it to another, but in doing so it had to move past a barrier.

“If you have more joints, instructions to the robot are complicated,” Quintero-Peña said. “If you’re directing a human, you can just say, ‘Lift up your hand.'”

But a robot’s programmers have to be specific about the movement of each joint at each point in its trajectory, especially when obstacles block the machine’s “view” of its target.

Rather than programming a trajectory up front, BLIND inserts a human mid-process to refine the choreographed options — or best guesses — suggested by the robot’s algorithm. “BLIND allows us to take information in the human’s head and compute our trajectories in this high-degree-of-freedom space,” Quintero-Peña said.

“We use a specific way of feedback called critique, basically a binary form of feedback where the human is given labels on pieces of the trajectory,” he said.

These labels appear as connected green dots that represent possible paths. As BLIND steps from dot to dot, the human approves or rejects each movement to refine the path, avoiding obstacles as efficiently as possible.

“It’s an easy interface for people to use, because we can say, ‘I like this’ or ‘I don’t like that,’ and the robot uses this information to plan,” Chamzas said. Once rewarded with an approved set of movements, the robot can carry out its task, he said.

“One of the most important things here is that human preferences are hard to describe with a mathematical formula,” Quintero-Peña said. “Our work simplifies human-robot relationships by incorporating human preferences. That’s how I think applications will get the most benefit from this work.”

“This work wonderfully exemplifies how a little, but targeted, human intervention can significantly enhance the capabilities of robots to execute complex tasks in environments where some parts are completely unknown to the robot but known to the human,” said Kavraki, a robotics pioneer whose resume includes advanced programming for NASA’s humanoid Robonaut aboard the International Space Station.

“It shows how methods for human-robot interaction, the topic of research of my colleague Professor Unhelkar, and automated planning pioneered for years at my laboratory can blend to deliver reliable solutions that also respect human preferences.”

Rice undergraduate alumna Zhanyi Sun and Unhelkar, an assistant professor of computer science, are co-authors of the paper. Kavraki is the Noah Harding Professor of Computer Science and a professor of bioengineering, electrical and computer engineering and mechanical engineering, and director of the Ken Kennedy Institute.

The National Science Foundation (2008720, 1718487) and an NSF Graduate Research Fellowship Program grant (1842494) supported the research.

Video: https://youtu.be/RbDDiApQhNo

Their findings were recently published in the International Society for Microbial Ecology Journal.

Humans rely on honey bees for food security. Because they will pollinate almost anything, honey bees are extremely useful for agriculture. But over the past few decades, the honey bee population has experienced dramatic declines caused by the effects of multiple stressors, the most pervasive of which is limited nutrition. Beekeepers in the United States reported losing 40.5 percent of their managed colonies between 2015 and 2016 alone, according to a national survey.

“The effects of poor nutrition are most damaging in the developing larvae of honey bees, who mature into workers unable to meet the needs of their colony,” said Irene Newton, a professor in the IU Bloomington College of Arts and Sciences’ Department of Biology, who led the study. “It is therefore essential that we better understand the nutritional landscape experienced by honey bee larvae.”

Newton said honey bees need to collect pollen and nectar from a variety of plants and flowers to help their colonies stay healthy throughout the year. But many bees in the U.S. lack this floral diversity.

“We’ve changed the way we use our land in the U.S.,” Newton said. “Now we have tons of monoculture crops like corn, which are wind pollinated and therefore no use to bees, covering acres and acres of land. Other crops that bees do pollinate are grown in monoculture as well, limiting the options for bees.

“If you limit yourself to only eating one thing, that’s not healthy for you. You have to have a broad diet that will help fulfill all of your nutritional needs. Bees are the same way.”

Honey bee larvae are fed by their sister bees. Their diet consists of foraged ingredients such as nectar and pollen, as well as royal jelly — a bee glandular secretion that is complex and nutrient rich. If larvae are destined to be queens, they will eat royal jelly their whole lives. If they are workers, their diets will shift to nectar and pollen after a few days.

In addition to being more nutritious than nectar and pollen, royal jelly has long been known to possess potent antimicrobial properties due to its acidity, viscosity and the presence of antimicrobial peptides. This means that most microbes exposed to royal jelly die, Newton said.

Except one.

According to their new study, Newton and her research team found that a specific microbe — Bombella apis — is the only larva-associated bacterium that’s actually able to thrive in royal jelly. They also found that B. apis makes royal jelly more nutritious by significantly increasing its amino acid content, which helps developing bees build resilience against nutritional stress.

“We have identified a nutritional symbiont of honey bees — a microbe that can help bolster the bees against nutrient scarcity and stress,” Newton said. “When we limited bee nutrition during development, we saw a drop in mass for the bees; bees were much smaller than their control counterparts.

“When B. apis was added to these same bees, although they had poor nutrition, they reached the same mass as control bees given full nutrition. The microbe was able to make up for the poor diet. This suggests that B. apis could be added to colonies as a probiotic to protect from nutritional stress.”

The results suggest that B. apis may have potential as a key supplement in future beekeepers’ efforts to counteract the negative influence of poor nutrition on honey bee health. B. apis can survive for over 24 hours in sugar water, so beekeepers who are already supplementing their colonies could potentially integrate a B. apis probiotic into their bees’ diets.

This research expands on over six years of previous studies by Newton and her colleagues, including findings that B. apis protects bees against fungal infections and is a significant part of the queen gut microbiome.

“We are excited to explore the other interactions that B. apis has in a colony, to better understand what it’s doing in different environments and the role it plays in association with honey bee queens,” Newton said.

To perform its myriad functions, the immune system must, above all, differentiate between self and non-self — a remarkable selective ability that allows it to detect and disable harmful agents while sparing the body’s own tissues.

If the immune system fails to make this distinction, it can mistakenly launch an assault against the body, causing autoimmune disorders.

Researchers have known the general principle underlying this selective ability for some time, but exactly how immune cells learn to distinguish friend from foe has remained less well understood.

Now, a new study led by researchers at Harvard Medical School identifies a new mechanism that explains how the body’s most powerful immune troops — T cells — learn to tell self and non-self apart.

The work, conducted mainly in mice, was published online June 16 in Cell and is scheduled to appear in the July 7 print issue.

The research shows that the thymus gland — the organ where T cells are born and trained — educates nascent immune cells by exposing them to proteins made by thymus cells that mimic various tissues throughout the body. Specifically, the research demonstrates that by assuming different identities, these specialized thymus cells preview for the maturing T cells self-proteins they would encounter once they leave their native thymus gland.

“Think of it as having your body recreated in the thymus,” said study senior author Diane Mathis, professor of immunology at Harvard Medical School. “For me, it was a revelation to be able to see with my own eyes muscle-like cells in the thymus or several very different types of intestinal cells.”

The findings, Mathis said, shed light on how the adaptive immune system acquires its ability to discern friend from foe. Glitches in this critical recognition system can have grave consequences.

“Our immune system is super powerful. It can kill any cell in our body, it can control any pathogen we encounter, but with that power comes great responsibility,” said study first author Daniel Michelson, an MD/PhD student at Harvard Medical School and a researcher in the Mathis/Benoist lab. “If that power is left unchecked, it can be lethal. In some autoimmune diseases, it is lethal.”

School for T cells

T cells, so named because they mature and learn to do their job in the thymus before they are released into the body, are the immune system’s elite forces charged with multiple functions. They recognize and eliminate pathogens and cancer cells; they form long-term memory of viruses and bacteria encountered in the past; they regulate inflammation and tamp down hyperactive immunity.

But how does a newborn T cell that’s never left the thymus know which proteins are the body’s own and which herald enemy presence?

“T cells get educated in the thymus, but the thymus is not a gut, it’s not a pancreas,” Michelson said. “There’s no reason why these T cells should be able to recognize these organs before they leave the thymus.”

Researchers knew that this early training does take place in the thymus, but the precise teaching tools the gland uses have eluded them.

A molecular explanation for a centuries-old observation

Until the mid-1900s, the thymus provoked little scientific interest because it was deemed vestigial, Michelson said. But as far back as the mid-1800s — well before scientists knew what the thymus does or that an adaptive immune system existed — biologists had already noticed cells in the thymus that looked out of place. Peering into their microscopes throughout the decades, they saw cells that looked like they came from muscle, intestine, and skin. Yet, the thymus was none of the above. The observations made no sense.

The newly published research hearkens back to a very old finding and puts it into a whole new molecular context, Michelson said.

The study showed that these teacher cells, dubbed mimetic cells for their ability to mimic different tissues, work by co-opting various transcription factors — proteins that drive the expression of genes unique to specific tissues. When they do so, the mimetic cells effectively adopt the identities of tissues such as skin, lung, liver, or intestine. They then present themselves to immature T cells to teach them self-tolerance, the team’s experiments showed.

The work shows that T cells-in-training that mistakenly react against self-proteins either receive a command to self-destruct or get repurposed into other types of T cells that don’t kill but instead restrain other immune cells from attacking.

“The thymus says: This cell is autoreactive, we don’t want it in our repertoire, let’s get rid of it,” Michelson said.

Plot twist

Until now, the elimination of self-reactive T cells was thought to be regulated largely by a single protein called AIRE. The Mathis/Benoist lab was critical in elucidating the function of AIRE. Defects in this protein can lead to a serious immune syndrome characterized by the development of multiple types of autoimmune disorders.

Mathis and Michelson went into their current research seeking to map the molecular pathways involved in AIRE function. Instead, they found many cells in the thymus that did not express the AIRE protein but were still capable of adopting the identities of different tissue types. AIRE, the researchers realized, was only part of the story.

The researchers say the newly identified mimetic cells are likely to play a role in various autoimmune diseases associated with the tissue types they mimic, a hypothesis they plan to pursue.

“We think it’s an exciting discovery that may open up a whole new vision of how certain types of autoimmune diseases arise and, more broadly, of the origins of autoimmunity,” Mathis said.

The researchers said their next steps are to acquire an even deeper understanding of the molecular mechanisms that underlie T cell education, to study the association between individual mimetic cell types and T cell function and dysfunction, and to determine how the mechanism plays out in the human thymus.

Study co-authors included Koji Hase of Keio University, Tsuneyasu Kaisho of Wakayama Medical University, and Christophe Benoist of HMS.

The work was funded by National Institutes of Health grants R01AIo88204, Ro1DKo60027, and T32GM007753.